A novel glove monitoring system for quantifying neurolog- ical symptoms …mediatum.ub.tum.de ›...
Transcript of A novel glove monitoring system for quantifying neurolog- ical symptoms …mediatum.ub.tum.de ›...
TECHNISCHE UNIVERSITÄT MÜNCHEN
Lehrstuhl für Mikrotechnik und Medizingerätetechnik
A novel glove monitoring system for quantifying neurolog-
ical symptoms during deep-brain stimulation surgery
Houde Dai
Vollständiger Abdruck der von der Fakultät für Maschinenwesen der
Technischen Universität München zur Erlangung des akademischen Grades eines
Doktor-Ingenieurs (Dr.-Ing.)
genehmigten Dissertation.
Vorsitzender: Univ.-Prof. Dr. med. Dr.-Ing. habil. Erich Wintermantel
Prüfer der Dissertation:
1. Univ.-Prof. Dr.rer.nat. Tim C. Lüth
2. Univ.-Prof. Dr.-Ing. Veit St. Senner
Die Dissertation wurde am 22.10.2013 bei der Technischen Universität München eingereicht und durch die Fakultät für Maschinenwesen am 12.05.2014 angenommen.
II
Danksagung
III
Danksagung
I have gained a profound amount of knowledge during my four years of study and research at
the Technische Universität München (TUM). Throughout these years, my research ability has
continually improved.
Here, I first wish to express my sincere gratitude to Prof. Dr. rer. nat. Tim C. Lüth, for his
constant support throughout my graduate studies at the Institute of Micro Technology and
Medical Device Technology (MIMED, TUM). I also wish to express my sincere gratitude to
the Chinese Scholarship Council (CSC) for their financial support of my studies in Germany.
Furthermore, I would like to thank Dr.-Ing. Lorenzo T. D'Angelo, for his patient instruction,
constant encouragement, and friendly assistance in relation to my studies and research. I
would like to thank Jakob Neuhäuser and Yan Zhao for their kind support, insightful research
discussions, and invaluable suggestions for this thesis. I also wish to express my sincere grati-
tude to Jiaxi Shi for his kind suggestions regarding the structure of my dissertation.
I would also like to offer my thanks to Dr. Jan H. Mehrkens, from Ludwig-Maximilians-
Universität München, for his expert suggestions in this study. I would also like to
acknowledge the great amount of help provided in this study by my student Bernward Otten.
In particular, many thanks to Jordan Evans, Jeremiah Hendren, and the other TUM English
Writing Center staff, for their proofreading of this thesis and their invaluable suggestions.
I would like to thank Mrs. Renate Heuser, Barbara Govetto, and Dr. Franz Irlinger for their
kind work.
Also, I offer my thanks to my colleagues Dr. med. Karin Tonn, Dr.-Ing. Khalil Niazmand, Ian
Somlai, Dr.-Ing. Axel Czabke, Cheng Fang, Joachim Kreutzer, and Samuel Reimer.
In addition, I would like to thank Professor Chao Hu, Dr. Wanan Yang, Zhenglong Chen, and
Professor Guotai Jiang, who have always offered their help.
I would like to thank all the kind people at MIMED and TUM.
Finally, to my family, I would like to thank them for their eternal support and encouragement.
Houde Dai,Oct.,2013
IV
Inhalt
V
Index
1. Introduction ...................................................................................................................... 1
1.1 Project Description ...................................................................................................... 1
1.2 Structure of the Thesis ................................................................................................. 2
2. Application Description ................................................................................................... 3
2.1 Clinical Need ............................................................................................................... 3 2.2 Technical Need ............................................................................................................ 7
3. State of the Art .................................................................................................................. 9
3.1 Neurological Symptoms and Assessment Tasks ......................................................... 9 3.2 Intraoperative Assessment of Neurological Symptoms ............................................. 13
3.3 Quantitative Assessment of Neurological Symptoms ............................................... 15
3.4 Commercial Systems for Quantification of Neurological Symptoms ....................... 19 3.5 Research Systems for Quantification of Neurological Symptoms ............................ 21 3.6 Inertial Sensors and Sensor Fusion for Motion Tracking .......................................... 24 3.7 Limitations of Existing Technology .......................................................................... 27
4. Glove Monitoring System .............................................................................................. 30
4.1 Task Description ........................................................................................................ 30
4.2 Expected Advantages ................................................................................................ 33
5. System Concept .............................................................................................................. 35
5.1 Designs with Wireless Communication Interfaces .................................................... 35
5.2 Static System Description .......................................................................................... 36
5.3 Dynamic System Description .................................................................................... 40
6. Prototypical Realization ................................................................................................ 51
6.1 Nine-Axis Direction-Cosine-Method Realization ..................................................... 51
6.2 Prototypes with Wireless Communication Interfaces ............................................... 52 6.3 Materials .................................................................................................................... 56
6.4 Physical Implementation ........................................................................................... 61 6.5 Graphical User Interface Implementation ................................................................. 67
6.6 Calibration of Inertial Sensors and Force Sensor Boxes ........................................... 68 6.7 Combined Version ..................................................................................................... 74 6.8 Conclusion ................................................................................................................. 76
7. Experiments and Discussion .......................................................................................... 78
7.1 Verification of Analytical Methods for Tremor Assessment .................................... 78 7.2 Verification of Hand Grasping Angle Calculation .................................................... 83 7.3 Verification of Analytical Methods for Rigidity Assessment ................................... 86
7.4 Experiment of Tremor Assessment ........................................................................... 98 7.5 Experiment of Bradykinesia Assessment ................................................................ 102 7.6 Conclusion ............................................................................................................... 104
8. Conclusions and Outlook ............................................................................................. 106
8.1 Conclusions ............................................................................................................. 106 8.2 Outlook .................................................................................................................... 107
9. Glossary ......................................................................................................................... 109
Index
VI
10. Bibliography .............................................................................................................. 112
VII
Introduction
1
1. Introduction
1.1 Project Description
Parkinson’s disease (PD) and essential tremor (ET) are the two common movement disorders,
which denote degenerative and progressive disorder of the central nervous system (CNS).
Tremor, bradykinesia, and rigidity are the three primary symptoms of PD (Elbe et al., 2011).
ET is a tremor of the hands when a patient is performing voluntary movements.
Deep-brain stimulation (DBS) is a crucial surgical procedure for PD, ET, and other neurologi-
cal symptoms (Okun et al., 2008). At present, DBS can only relieve the severe symptoms of
PD, ET, and other nervous system disorders, and then improve the quality of patients’ lives.
The precise mechanisms of PD and how DBS works remain uncertain at present (Dauer et al.,
2003). However, the stimulation in specific areas of the patient’s brain by sending high fre-
quency electrical impulses can alleviate symptoms or diminish the side effects of medications.
Therefore, DBS is the most effective surgical procedure for patients with severe symptoms of
PD, ET, dystonia, or dyskinesia when the patients are insensitive to drug medications.
The positioning target area of an electrode is first confirmed by the three-dimensional struc-
ture obtained from magnetic resonance imaging (MRI). During the positioning of the elec-
trode in the brain by using microelectrode-guided mapping, the neurosurgeon may choose an
optimal location based on the results of sensorimotor mapping or subjective methods, involv-
ing observation according to the clinical scales (Unified Parkinson’s Disease Rating Scale,
etc.) and the assessment of handwriting such as a drawing of an Archimedes spiral. This is
usually accompanied by several small movements of the electrode in a target area. Although
used under clinician observation, these largely subjective scales lack validation against the
actual stimulating effect. Furthermore, the coarse resolution of the ratings is insufficient for
assessing small changes in symptom severity. Finally, the extent of inter-clinician and inter-
subject rating variability is unknown (Machado et al., 2003).
There is no designated instrumental method for the accurate monitoring of the stimulating
effect during DBS surgery as of yet. Nevertheless, neurosurgeons need to know the exact po-
sition and stimulating intensity of the deep-stimulation electrode to achieve the best effect.
The goal of the present project is to develop a portable assessment system to quantify the
three primary symptoms of PD and ET during DBS surgery. These movement disorders and
their changes should be monitored to obtain the optimal electrode target position and stimula-
tion intensity during intraoperative stimulation. It is also supposed to support choosing the
optimal stimulation settings of the DBS electrodes after DBS surgery.
In this study, a new concept for quantifying neurological symptoms during DBS surgery was
developed, specifically concerning portability, wearability, and intraoperative application. In
this concept, the current possibilities of inertial sensor technology and motion-tracking algo-
rithms could be used to implement quantitative assessments of the primary neurological
symptoms without disrupting existing DBS processes. User-friendly human-machine interfac-
es were designed to monitor the changes of symptoms.
Introduction
2
1.2 Structure of the Thesis
The organization of this thesis is as follows.
Chapter 2 describes the motivation of this study, both for clinical and technical needs. The
introduction of PD, ET, and DBS are presented in this chapter.
Chapter 3 reviews the intraoperative assessment approaches of neurological symptoms and
the systems that can be used to quantitatively assess these neurological symptoms. Inertial
sensors and sensor fusion methods for motion tracking are also introduced in this chapter.
Chapter 4 describes the parameters, goals, and expected advantages of the glove monitoring
system. Its system concepts are presented in Chapter 5, which includes static and dynamic
system descriptions. The quantification algorithms of neuromotor symptoms are based on
statistical analyses, especially the estimation theory and its least squares methods.
For the prototypical realization in Chapter 6, designs with wireless communication interfaces
are presented first. In the first step of this study, the glove monitoring system was divided into
two separate systems: one for tremor and bradykinesia assessment and the other for rigidity
assessment. Their prototypes were based on six-axis motion tracking sensors (gyroscope and
accelerometer) and force sensors. According to the operations during clinical experiments and
the suggestions of surgeons, a modified version with combined hardware and software was
developed. As a result, a prototype that could assess all three symptoms was implemented.
Chapter 7 describes the verification of analytical methods, and clinical experiments. Com-
parative experiments between these two prototypes and an electromagnetic motion-tracking
system were carried out. The prototype for tremor and bradykinesia assessment was tested
with clinical experiments in the hospital.
Finally, conclusions and future work are given in Chapter 8.
State of the Art
3
2. Application Description
The information about PD, ET, and DBS surgery are presented in Chapter 2.1. The technical
need to quantify neurological symptoms is introduced in Chapter 2.2.
This chapter shows that the quantification of neurological symptoms during DBS surgery
plays a key role in the surgical treatment of PD or ET. It provides feedback to the stimulation
settings of each DBS electrode. Thus the optimal location for the electrode implantation can
be found.
2.1 Clinical Need
2.1.1 Parkinson’s Disease and Essential Tremor
It was estimated that 1% of 70-year-olds suffer from PD (Chaudhuri et al., 2011). About 10%
of PD patients are younger than 50 years old. At present about four to six million people are
PD patients. A series of studies indicated the overall prevalence of ET to be 0.4%–0.9% and
the prevalence in 60-year-olds to be 4.6% (Louis & Ferreira, 2010). Due to demographic in-
crease of the elderly population, movement disorders such as PD and ET will occur more fre-
quently in the future.
Figure 2-1: Symptoms of Parkinson’s disease (Taken from Gowers, 1886). Parkinson’s disease affects
patients in many different ways with a variety of symptoms. These symptoms can be classified as motor
symptoms, neuropsychiatric symptoms, and autonomic dysfunction.
PD occurs mainly in a patient’s hands, feet, and head. As shown in Figure 2-1, tremor
(rhythmic back and forth motion), bradykinesia (slowness of motion), rigidity (resistance to
movement), poor balance, and parkinsonian gait are the five motor symptoms of PD
(Jankovic, 2013). Poor balance and parkinsonian gait, however, are hard to assess during DBS
surgery. Tremor is a central symptom of PD, and is generally judged according to hand tremor
with a particular frequency (3.5–7.5 Hz) and amplitude (speed and range) (Salarian et al.,
2004). Bradykinesia is a feature of basal ganglia disorders. It involves difficulties with plan-
ning, beginning, and executing movement; and with performing sequential and simultaneous
tasks. Rigidity means increased muscle tone, which is defined as a resistance to a passive
movement.
State of the Art
4
An ET can be difficult to distinguish from a parkinsonian tremor. However, ET generally pre-
sents itself with a higher frequency (4–12 Hz) and occurs only when the affected muscle is in
an active state. Physical or mental stress will also make ET worse. Therefore, rest tremor is
usually not part of the ET. In addition, the target areas of an electrode for ET and
parkinsonian tremor are different. The assessment method for both ET and parkinsonian
tremor, however, can be almost the same (Fekete et al., 2010).
Problem: Reduce the Severity of Symptoms
The mechanism responsible for PD and ET is not clear now. The primary goal in PD and ET
treatment strategies is to maintain a patient’s independence and quality of life, and at the same
time minimize potential complications of treatment. The challenge of the PD and ET treat-
ment strategies is to reduce the severity of symptoms.
At first, PD patients are treated with Levodopa medications. However, drug efficacy decreas-
es in later stages, together with the period of drug medication. For the patients with severe
symptoms of PD or ET, who are insensitive to medications, DBS is the most effective treat-
ment (Kringlbach et al., 2007).
2.1.2 Deep-Brain Stimulation Surgery
DBS was approved by the Food and Drug Administration (FDA) as a surgical treatment for
ET in 1997 and for PD in 2002. DBS can only ease some symptoms of a patient with PD or
other neurological disorders. Medications for the patient are needed even after DBS surgery
(Bittar et al., 2005). More than 80 000 DBS surgeries had been performed worldwide up until
2011 (Oluigbo et al., 2012).
Figure 2-2: DBS system (© Medtronic, 2009). Left: DBS system diagram; middle: DBS clinician pro-
grammer (size: 22cm × 10cm × 4cm); right: DBS patient programmer (size: 9.4cm × 5.6cm × 2.8cm).
Meditronic Inc. is the major manufacturer of DBS devices.
State of the Art
5
A deep-brain stimulation system includes neurostimulators (a surgically implanted, battery-
operated medical device), electronic leads (also called electrodes, two thin, insulated wires for
each side of the brain), and extensions. These components can be placed in either one or both
sides of the brain. They are implanted inside the body and powered by batteries. DBS uses a
neurostimulator, which is similar to a heart pacemaker and approximately the size of a stop-
watch, to deliver electrical stimulation to targeted areas in the brain that control movement,
thus blocking the abnormal nerve signals that cause motor symptoms. As Figure 2-2 shows,
there are also two programmers which can be used by the surgeon or the patient outside the
human body. The stimulating mode, intensity and frequency of the neurostimulators can be
adjusted using the clinician programmer. The patient programmer can be used to turn the
therapy on or off.
Current Application Flow of DBS Surgery
At first, a stereotactic head frame is placed on the patient’s head to precisely target the brain
structure. The neurosurgeon uses magnetic resonance imaging (MRI) or computed tomogra-
phy (CT) scanning to identify and locate the exact area (coordinate) within the brain in which
electrical nerve signals generate the PD or ET symptoms. A three-dimensional (3D) offline
graphics program provides detailed information on the function of the area (Kringlbach et al.,
2007).
Then the patient enters the operating room and lies on a surgical bed in a reclined and com-
fortable position. The head frame is fixed to the table. In general, at least five members of the
hospital staff are required in the operating room, including a surgeon, a nurse, an anesthesiol-
ogist, a technician, and an assistant, during DBS surgery. The installation of the components
is under the condition of general anesthesia (Bittar et al., 2005).
As Figure 2-2 shows, a surgeon has drilled two small holes in the patient’s skull and has
threaded two insulated wires, with electrodes attached at the top of the brain. Lead wires are
run under the skin from the brain to the upper chest. The lead wires are attached to two
neurostimulators that are surgically implanted in the chest and provide electricity to stimulate
the brain. Two electrodes and neurostimulators on both sides are implanted one by one by
using the same surgical procedure.
Each neurostimulator sends electrical pulses to one of the three active areas of the brain via
the electrode to interfere with neural activity. There are a few target sites inside the brain for
achieving differing results, and the most common sites are the subthalamic nucleus (STN) and
the globus pallidus interna (GPi). In addition, the caudal zona incerta and the pallidofugal
fibers medial to the STN are being assessed in some situations. Then a site will be chosen for
each electrode based on the individual patient (Bittar et al., 2010).
Challenge: Select an Optimal Electrode Position
An optimal electrode position is significant for the effect of surgery. The implantation of elec-
trodes is performed only one side at a time. Before the surgery, the coordinates for the elec-
trode target area inside the brain, which can be seen from Figure 2-3, are identified with the
help of an optimum radiologic targeting (MRI-based) and a 3D offline graphics program.
The key point of the surgery is to define an optimal point for each electrode positioning.
For each electrode, 4 to 10 points inside the brain are tested during the surgery in order to
evaluate the electrode’s position and stimulation intensity. This is usually accompanied by
State of the Art
6
small movements in a small target area. After surgery, a neurologist also needs to adjust the
stimulation amplitude of the stimulator which is implanted in the chest of the patient for the
best treatment effect (Machado et al., 2006).
Figure 2-3: Electrode implantation (Based on Hotzheimer & Mayberg, 2012). The tips of the electrodes
are implanted one by one within the targeted brain area.
During DBS surgery, it is important to monitor the patient’s stimulation effects, based on his
or her individualized response to the stimulation. The quantified severity values of symptoms
can be used as feedback for the stimulating setting of the DBS electrode (Kern et al., 2007).
After its optimal target is found, an electrode is permanently implanted. For the electrode with
misplacement or in a suboptimal target, the positive effect of DBS cannot be reached or re-
stricted. Some patients need to adjust the electrode through another DBS surgery.
Assignment: Monitor the Severity of the Neurological Symptoms
Some surgeons may use microelectrode recording (MER), which involves a small wire that
monitors the activity of nerve cells in the target area, to more specifically identify the precise
brain target that will be stimulated.
The current standard for evaluating motor symptoms associated with PD is the Unified Par-
kinson’s Disease Rating Scale (UPDRS), a qualitative assessment that is completed by the
subjective judgement of the surgeons. Essential Tremor Rating Assessment Scale (TETRAS)
is the standard for evaluating the severity of ET. Motor symptoms are rated on a scale from 0
to 4 corresponding to normal, slight, mild, moderate, and severe (Tagliati et al., 2007).
The Hoehn and Yahr scale, Schwab and England Activities of Daily Living Scale, and Web-
ster Scale are also used for the assessment of PD. These subjective assessment ratings lead to
problems when evaluating the effectiveness of therapies for PD or ET (Pahwa et al., 2007).
At present, there is no designated instrumental method of monitoring the immediate motor
effects of DBS procedure. Nevertheless, surgeons need to get feedback on the stimulating
electrode in order to achieve optimal therapeutic efficacy. There is, therefore, a need to de-
velop a monitoring system that is able to quantify tremor, bradykinesia, and rigidity in real-
time during DBS surgery, instead of subjective assessment.
Target area
Electrode
State of the Art
7
A DBS stimulator can be turned on and programmed in three to four weeks after implantation.
The first programming session can take several hours in order to ensure the device is correctly
functioning and that various stimulation parameters (voltage, frequency of the DBS stimula-
tor, and electrodes mode) are optimally programmed.
Routine outpatient programming sessions at approximately one, two, and four months after
DBS surgery are needed for the patient. Because the neurostimulator contains a battery, it
must be surgically replaced every three to five years.
2.1.3 Special Concerns - Motor Fluctuations and Dyskinesia
About 50% of PD patients have motor fluctuations (MF) and dyskinesia after five years of
levodopa medications. As Figure 2-4 shows, motor fluctuations indicate the effective period
of certain doses is shorter all the time (this is known as “end-of-dose deterioration”). It also
means the alterations between “ON”, which is a state of good response to anti-parkinsonian
medications, and “OFF”, which is a state for patients experiencing pakinsonian symptoms.
The symptoms of a PD patient may reappear unexpectedly and quickly. This switch sensation
is described as “ON/OFF” syndrome. Levodopa-induced dyskinesia (LID) involves a series of
hyperkinetic movements (involuntary, episodic, and irregular) such as athetosis, chorea, and
dystonia (Hause et al., 2000).
Figure 2-4: Motor fluctuations and dyskinesia. Levodopa’s therapeutic window narrows overtime during
ON state. As a result, the patient’s response to Levodopa shortens over time.
Dyskinesia is one of the side effects of DBS surgery. Dyskinesia appears when the electrode
stimulating voltage is too high. Thus, it is important to avoid such over-stimulation.
2.2 Technical Need
In 1872, Jean-Martin Charcot (1825–1893), the most celebrated clinical neurologist of the
19th century, developed a tremor recording device, as Figure 2-5 shows. This device provided
a new approach to assess the symptom severity of movement disorders, other than visual ob-
servation and clinical maneuvers.
At present, several groups have used electromyography (EMG), computer tracking, digital
tablets, infrared video cameras, and laser transducers to assess tremor and bradykinesia objec-
tively. All these solutions have demonstrated limited usability in clinical settings due to defi-
ciencies in wearability, fidelity, and flexibility (Pahwa & Kvons, 2007).
ON state without LID
Peak-dose dyskinesias
Peak-dose dyskinesiasOFF state
Levodopa Administration (Date)
ON state with LID
Levodopa C
oncentr
ation (
Dose)
OFF
ON
ON &Dyskinesia
State of the Art
8
Figure 2-5: Charcot’s early tremor recording device (lower left) and its recordings (lower right). Charcot
used the sphygmograph to record tremors in the wrist. The resultant tremor recordings were conducted
at rest (AB) and during activity (BC) of a patient with Parkinson’s disease (Based on Pahwa & Kvons,
2007).
There has recently been growing interest in the application of body-fixed sensors (BFS) and
in particular inertial sensors for long-term monitoring of PD symptoms. Micro-Electro-
Mechanical Systems (MEMS) gyroscope and accelerometer are widely used for the motion
tracking in tremor and bradykinesia assessments. With the development of new MEMS tech-
nology, the dimension of the sensor circuit board is smaller and the signal processing is easier
to be carried out than before.
Other kinematic sensors and force sensor have been used to detect and quantify rigidity.
These sensors were placed on the body, feet or arms of the patients and were not specifically
designed for DBS (Patel et al., 2008).
The assessment system of neurological symptoms for DBS surgery must strictly adhere to the
requirements in the operation room. It is important to perform three assessment tasks in a
portable and wearable system with small dimensions. This system should be safe and not re-
strict the patient’s movement.
In addition, a user-friendly human-machine interface should be designed to support the sur-
geons to monitor the changes of symptoms.
A
A
A
B
B
B
C
C
C
State of the Art
9
3. State of the Art
The clinical assessment tasks of tremors, bradykinesia, and rigidity is first introduced in
Chapter 3.1.
The intraoperative assessment of neurological symptoms is introduced in Chapter 3.2. At pre-
sent there are only two intraoperative approaches to assess neurological symptoms:
MER/EMG techniques and subjective assessment by the surgeons.
There are also some systems which are used to quantitatively assess neurological symptoms
outside of the operating room. Some researchers and companies have tried to assess the sever-
ity of neurological symptoms and their changes (OFF and ON state).
The quantitative assessment methods of neurological symptoms in these systems are intro-
duced in Chapter 3.3. In addition, these systems are classified into commercial systems and
systems in research and development. Detailed information of the assessment systems is de-
scribed in Chapter 3.4 and Chapter 3.5 respectively. However, most of them are for tremor
and bradykinesia assessment. Rigidity assessment is available only in a few research projects.
MEMS inertial sensors and sensor fusion algorithms for motion tracking are introduced in
Chapter 3.6.
Limitations of the existed technology (disadvantages of the state of the art) are described in
Chapter 3.7.
3.1 Neurological Symptoms and Assessment Tasks
In clinical practice, tremor assessment tasks include rest tremor, postural tremor, and action
tremor movements. Bradykinesia measurement tasks include finger tapping, hand grasping,
and rapidly alternating movements. Passive elbow or wrist movements (repeatedly flexing
and extending) are used for rigidity assessment.
A detailed instruction of the assessment tasks is described as follows.
3.1.1 Tremor Assessment
Symptom
Tremor syndromes (oscillatory movements) are classified into three primary types: rest trem-
or (RT), postural tremor, and action tremor. Rest tremor, which is the characteristic of
parkinsonian tremor, happens when a body part is relaxed, for example, when lying in bed.
Postural tremor, which is the characteristic of ET, occurs while a body part is maintaining a
position against gravity. Action tremor happens when a voluntary contraction of a muscle,
follows the condition, for example, holding a cup. In most situations, parkinsonian tremor
manifests the combination of rest, postural, and action tremors (Deuschl et al., 1998). ET oc-
curs in postural tremor and action tremor.
Parkinsonian tremor is the central symptom of PD. It is prominent in PD, about 70% of PD
cases. The tremor associated with PD has a characteristic appearance, and reduces with pur-
poseful activity. Symptoms often start with an occasional tremor in one finger that spreads
State of the Art
10
over time to involve the whole arm. ET is present when the limb is at rest or held up in a stiff
unsupported position, and usually disappears briefly during movement.
Assessment
As shown in Figure 3-1, when the rest tremor is assessed, the patient should sit quietly in a
chair or lie down in bed with his or her hands and feet comfortably placed for several seconds
with no other activity. The stable tremor amplitude gives the final score.
a) b) c)
Figure 3-1: Tremor assessment tasks: a) rest tremor; b) postural tremor; c) action tremor. When lying in
the bed during DBS surgery, the assessment tasks for the patient are the same to the general clinical ex-
ams (Based on Dai et al., 2013a).
When testing postural tremor, an examiner instructs the patient to stretch his or her arms out
with the palms facing down. The wrist should be straight and the fingers comfortably sepa-
rated so that they do not touch each other. This posture is observed for ten seconds.
Action tremor of the hands is tested by the finger-to-nose maneuver or by holding a cup. With
the arm starting from the outstretched position, the patient performs at least three finger-to-
nose maneuvers with each hand reaching as far as possible to touch his/her nose or the exam-
iner’s finger. The finger-to-nose maneuver should be performed slowly enough so as not to
hide any tremor that may occur with very fast arm movements. This is repeated with the other
hand, with each one rated separately. The tremor may be present throughout the movement,
and the highest stable tremor amplitude is rated.
The patient needs to completely relax during the rest tremor task because the tremor may be
heightened by the mental load. The patient should outstretch two arms during a postural
tremor task. Drinking, spoon, and spiral movements can be chosen as additional action tremor
tasks.
3.1.2 Bradykinesia Assessment
Symptom
Bradykinesia often initially manifests as slowness in performing activities. It is the most char-
acteristic feature of PD and is often associated with an impaired ability to adjust the body’s
position (Jankovic et al., 2008).
Bradykinesia encompasses slowness, decreased movement amplitude, and dysrhythmia. It
means the inability of generating maximum speed, power, or force. According to the modified
bradykinesia rating scale (MBRS), the speed, amplitude, and rhythm of bradykinesia task rep-
resent the parameters of bradykinesia, hypokinesia, and dysrhythmia, respectively.
Hypokinesia refers to a decreased amplitude or range of bodily movement. In addition, the
difficulty in selecting or activating motor programs in the CNS may result in akinesia (inabil-
State of the Art
11
ity to initiate movement) in the patient’s daily life. Akinesia (absence of movement) during
bradykinesia task is the delay (action time) of the patient to start the assessment task after the
instruction from the examiner.
Assessment
Bradykinesia measurement tasks include finger tapping, hand grasping, and rapid alternating
movements (Heldman et al., 2008).
Figure 3-2: Finger tapping.
As shown in Figure 3-2, finger tapping is used to assess bradykinesia. The subject is in-
structed to tap his/her fingers in rapid succession as quickly and as widely as possible. After
five seconds of practice, such open-close movement is performed for several seconds or sev-
eral times. Speed, amplitude, halts, and any decline in amplitude are evaluated. At least two
contact sensors on two fingers are used for the measurement of tapping duration. In order to
measure the range of the finger tap actions, two gyroscopes and two accelerometers are also
used when finger tapping is adopted.
Figure 3-3: Supination-pronation movements of the hand.
As shown in Figure 3-3, supination-pronation movement is also used to assess bradykinesia.
The patient is instructed to extend the arm out in front of his/her body with the palms facing
down and then to turn the palm up and down alternately several times (or several seconds) as
fast and as fully as possible.
Figure 3-4: Whole-hand grasping (Based on Dai et al., 2013a).
State of the Art
12
Whole-hand grasping is another movement used to assess bradykinesia. Figure 3-4 shows the
bradykinesia task, when all the fingers are closing and opening repeatedly. A single closing
and opening movement is regarded as a grasp cycle. The subject is required to grasp with the
greatest possible range and frequency. A single assessment task lasts for 10 to 15 seconds.
For supination-pronation and whole-hand grasping movements, a triple-axis gyroscope is
necessary to measure the angular displacement of the hand or finger. The gyroscope signals
obtained from patients with mild bradykinesia have a consistent amplitude and frequency and
appear sinusoidal. However, signals from patients with severe bradykinesia have much lower
and inconsistent amplitude and frequency. Speed, amplitude, halts, hesitations, and any de-
cline in amplitude are evaluated.
3.1.3 Rigidity Assessment
Symptom
Rigidity responds immediately upon PD treatment. It refers to a permanently elevated muscle
contraction, independent of passive movement velocity. Patients with severe rigidity can
hardly reach muscle relaxation and their voluntary movements are accompanied by an ele-
vated contraction of antagonist muscles (Shapiro et al., 2007).
Cogwheel rigidity, which means the muscles perform ratchet jerks when passive force bends
the limb, always appears as an early sign of PD. It performs a cogwheel mechanism with the
frequency range from 6 to 9 Hz and depends on the stretch velocity. This phenomenon seems
to be the combination of rigidity and superimposed tremor, which has a higher frequency than
parkinsonian tremor.
Assessment
Hand rigidity is more difficult to measure, and there is currently no standardized objective
method for measuring rigidity. Quantification of the mechanical properties of a joint can be
realized by passive joint movement, for example, flexion and extension of the joint by a clini-
cian or a torque motor. Rigidity is commonly assessed in the upper limbs at the wrist or el-
bow.
Figure 3-5: Elbow flexion-extension. An examiner holds the elbow of the patient for passive flexion-
extension movement. The examiner first puts one hand on the fixpoint. At the same time, his/her other
hand holds the patient’s wrist, on both sides, to perform flexions and extensions of the patient’s elbow.
As Figure 3-5 shows, passive flexion and extension of the elbow is used to assess rigidity.
The examiner flexes and extends the elbow joint through the force at the wrist. In clinical
practice, each test lasts for 10 to 30 seconds.
Fixpoint
Force point
Force point
State of the Art
13
3.2 Intraoperative Assessment of Neurological Symptoms
Subjective Assessments
The assessment of rest tremor (RT), action tremor, and postural tremor in the clinic is current-
ly mainly based on subjective methods and MER techniques according to clinical scales
(UPDRS and TETRAS) (Elbe et al., 2010).
When assessing bradykinesia during the DBS surgery, a neurologist asks the patient to per-
form rapid, repetitive, and alternating movements of the hand. The slowness level of the mo-
tion is scored according to the UPDRS. Based on experiences, the examiner classifies
bradykinesia severity on a four-point scale from 0 to 4 (Salarian et al., 2007).
In clinical practice, rigidity assessment is realized through passive movement of the subject’s
limb, which is controlled by a neurologist or another examiner. The level of instinctive re-
sistance to the exerted movement is scored according to the UPDRS ratings. Based on experi-
ences, the examiner classifies rigidity on a scale from 0 to 4, which is compared to a control
group (Patrick et al., 2001).
Quantitative Assessments: EMG and MER
At present, the Electromyography (EMG) method is used to measure tremor during DBS sur-
gery in some hospitals. The EMG recording technique allows the detection and monitoring of
electrical muscle activity following the attachment of surface electrodes to the surface of se-
lected muscles (Rissanen et al., 2007).
Figure 3-6: ISIS MER system (© inomed, 2012).
In addition, intraoperative microelectrode recording (MER) facilitates the surgeon in targeting
the optimal placement of each electrode. The surgeon connects small microelectrodes into the
intended target area and observes the pattern of neuronal activity to physiologically confirm
the optimal stimulating position of the electrode. The small tips of the microelectrodes are
placed close to the DBS electrode. They acquire the electrode activity (5100 µA) of individ-
ual neurons at a very high frequency (300 Hz) (McClelland et al., 2011).
Figure 3-6 shows the ISIS MER system with additional EMG function from inomed GmbH,
Germany. It includes a monitor, operating panel, main isolation unit, computer, printer, loud-
State of the Art
14
speaker, ISIS headboxes, ISIS neurostimulator, keyboard, mouse, and a drawer for accesso-
ries.
Figure 3-7: Electrode implantation and intra-operative MER recording during DBS surgery at the Uni-
versity Hospital of Munich (LMU) (© Mehrkens LMU, 2011). a) electrode implantation; b) MER graph-
ical user interface.
Figure 3-7 shows the five-channel MER recording during a DBS surgery at the University
Hospital of Munich (LMU, Germany). At the same time, there was an audio feedback of the
relative neuronal activity.
Because there are differences between different patients’ brains, the information obtained
from MER provides a more accurate target as final DBS placement than the information from
the MRI figures. Surgeons and technicians visualize and hear the neuronal activity from dif-
ferent points of the target area to identify specific structures based on the unique patterns of
neuronal activity (Bittar et al., 2005).
1
2
3
1) Surgeon
2) Sterotactic frame
3) Patient
Sound button MER signal
a)
b)
State of the Art
15
At the same time, the surgeon may move the patient’s joints or ask the patient to perform
physical examinations. The action of holding a cup or other objects by the patient is usually
assessed to help detect hand tremor. Fast hand movement is adopted for bradykinesia assess-
ment. Passive movement of the patient’s joint is used for rigidity assessment. The surgeon
assesses the symptom severity according to the clinical ratings based on experience.
The technician working with the MER or EMG equipment records the level of symptom se-
verity when the electrode, after being implanted within the target brain area, is slowly moving
within the brain. The surgeon can precisely map the target area (sensorimotor portion) in this
way and the optimal location of the electrode is identified (Rissanen et al., 2011).
Interventional MRI
Recently, an interventional MRI (or iMRI) with real-time imaging has appeared that has a
higher target accuracy than the MER system during the surgical procedure, which as a result
shortens the procedure time, even without a MER system (Ostrem et al., 2013). However, as
it is available in only a few hospitals, we do not discuss it in this study.
3.3 Quantitative Assessment of Neurological Symptoms
The summaries of quantitative assessment methods of tremor, bradykinesia, rigidity are listed
from Table 3-1 to Table 3-3, respectively. The quantitative assessment method of dyskinesia
is introduced in Chapter 3.3.4.
3.3.1 Tremor Quantification
Figure 3-8: Representative segments of the normalized time series of tremor signals (Based on Timmer et
al., 2000a). a) parkinsonian tremor; b) essential tremor. These data were recorded from the main direc-
tion (up and down) of the tremors by the usage of a single axis piezoresistive accelerometer. These types of
sensor data are usually sufficient for the analysis of ET but are insufficient for parkinsonian tremor be-
cause parkinsonian tremor is the movement of more than one dimension.
0 2 4 6 8 10
-2
-1
0
1
2
Acce
lera
tio
n (
g)
Time (s)
0 2 4 6 8 10
-1
0
1
2
3
Acce
lera
tio
n (
g)
Time (s)
a)
b)
State of the Art
16
Tremor amplitude and frequency (3.5–7.5 Hz for parkinsonian tremor; 4–12 Hz for ET) are
the important features (Elble et al., 2011).
Tremor is the most apparent and well-known symptom of PD. According to the patient data
from Prof. Dr. Jens Timmer at Freiburg University, representative segments of normalized
time series of essential tremor and parkinsonian tremor are shown in Figure 3-8.
Timmer et al. revealed that both parkinsonian tremor and ET exhibit a 2-order nonlinear os-
cillation, which is not strictly periodic (Timmer et al., 2000).
Table 3-1: Summary of tremor assessment methods. Peak power represents the power estimation
around the dominant frequency in the power spectrum of sensor signals; RMS refers to the root mean
square value; and “−” denotes no exact information.
Company or
author
Tremor Assessment Systems
Device or
system Sensors
Sensors
position Amplitude parameter
– Examiner Sight of the sur-
geon – UPDRS score
Inomed GmbH ISIS MER MER needle Brain Neuronal activity (voltage)
Giuffrida et al.,
2009 Kinesia
EMG,
gyroscope,
accelerometer
Finger Peak power
Burkhard et al.,
2002 MOTUS Gyroscope Palm Peak power
Narcisa et al.,
2011 CATSYS Accelerometer Hand RMS of accelerations
Patel et al.,
2009 Shimmer
Gyroscope,
compass Body Data range
Synnott et al.,
2012 WiiPD
Accelerometer,
infrared Hand RMS of accelerations
Salarian et al.,
2007 ASUR Gyroscope Wrist RMS of angular velocities
Spyers-Ashby
& Stokes, 2000 FASTRK
Electromagnetic
sensor Hand RMS of displacement
An overview of recent approaches in tremor assessment is given in Table 3-1. These sensor-
based systems could access the alternate between therapy “OFF” and “ON” states during
stimulation and medication. Most tremor assessment methods were based on the inertial sen-
sors (gyroscope and accelerometer) and a computer-based system. Sensors were placed on the
body, feet or arms of the patient.
More specifically, quantification of tremor has been achieved by numerical methods such as
time-domain analysis, spectral analysis, time-frequency analysis, and nonlinear analysis.
According to the research by Elble et al. (2006), which enrolled 928 patients, hand tremor
amplitude is logarithmically related to the 5-point clinical tremor ratings. According to the
research by Giuffrida et al. (2009), for the rest and postural tremor in PD, the summation of
logarithm peaks in both the power spectrums of accelerometer and gyroscope data has the
highest correlation with UPDRS scores, while the coefficient of determination (r2) was
around 0.9. For the action tremor in PD, the root-mean-square (RMS) sum of both gyroscope
State of the Art
17
and accelerometer data has the highest correlation with UPDRS (r2 = 0.69) (Mostile et al.,
2010).
3.3.2 Bradykinesia Quantification
Bradykinesia is quantified through finger tapping, whole-hand grasping, and supination-
pronation movements of hands.
Several groups and companies have used computer tracking, digital tablets, infrared video
cameras, or laser transducers in order to objectively assess parkinsonian bradykinesia.
An overview of recent approaches in bradykinesia assessment is given in Table 3-2.
Table 3-2: Summary of bradykinesia assessment methods. Here “–” denotes no exact information.
Company or
author Joint/ Task
Angle
measurement Parameters
– Examiner Sight of the sur-
geon –
Inomed
GmbH ISIS MER MER needle Neuronal activity (voltage)
Niazmand et
al., 2011a
Fingers/
Finger taps
Accelerometer,
touch sensors
Average and standard
deviation of the duration
Salarian et al.,
2007
Wrist/
Postural move Gyroscope Speed, amplitude, rhythm
Heldman et
al., 2011
Hand/ Tap, grasp&
pronation-supination
Accelerometer,
gyroscope, EMG Speed, amplitude, rhythm
Kim et al.,
2011a Fingers/ Finger taps Gyroscope Root-mean-square, cycle
Su et al.,
2003 Hand/ Grasps
Electromagnetic
sensors Speed, frequency
Kim et al. (2011a), from Konkuk University, Korea, quantified parkinsonian bradykinesia
during finger taps using a gyroscope. RMS velocity, RMS angle, and the estimated power
around dominant frequency were correlated well with clinical finger tapping scores.
3.3.3 Rigidity Quantification
For the quantitative assessment of parkinsonian rigidity, there are no available devices on the
market. According to research conducted by Patrick et al. (2001), expense, complexity, and
time involved are the most common reasons for not introducing quantitative rigidity evalua-
tion in clinical praxis.
Some research projects have tried to explore the relationship between biomechanical parame-
ters and the UPDRS rigidity scale. In most research, an examiner or a motor drive flexes and
extends a joint repeatedly, and then parameters from the applied torque are calculated. How-
ever, there are also researchers who calculate rigidity parameters from the electromyographic
potentials during flexion-extension movement. Patrick’s research indicated that the correla-
tion of mechanical properties with the UPDRS scores is superior to the correlation of EMG
with the UPDRS scores (correlation coefficient r: 0.60–0.86 compared to 0.37–0.79) (Sakoda
et al.; Patrick et al., 2001).
State of the Art
18
Table 3-3: Summary of rigidity assessment methods. Here “+” denotes the system performance in a
certain aspect, the amount of “+” represents a better performance; and “–” denotes no exact infor-
mation available.
Methods Assessed
joint
Inter-
variability
Intra-
variability Parameter
Signal
process Size
Fixed
frequency
Device or
reference
Feeling Elbow + ++ Feeling of
examiner – – No
Examiner
test
Electrical
current Brain +++ +++
Neuronal
activity + + – ISIS MER
Electrical
current
Hand/
Foot +++ +++
Muscle
activity + + – EMG
EMG;
torque-
angle
Wrist +++ +++ Work ++ + Yes Shapiro et
al., 2007
Torque-
angle Elbow ++ +++
Mechanical
impedance +++ +++ No
Patrick et
al., 2001
Torque-
angle Elbow +++ +++
Viscoelastic
values ++++ ++ Yes
Park et al.,
2011
Force-
angle Elbow ++ +++
EMG,
torque bias ++++ +++ No
Endo et
al., 2009
Torque-
angle Elbow ++ +++
Viscoelastic
values ++++ ++ No
Prochazka
et al., 1997
Some systems are designed to model the relationship between changing joint angles (degree)
and measured torque (N·m), which includes non-neural torque and neural torque. Force or
torque transducer, EMG, and position or angle sensors are used. Some approaches use kine-
matics to restrict the movement of the limbs, while others do not. An overview of recent ap-
proaches is given in Table 3-3.
A potentiometer is easy to use for angle estimation. However, a potentiometer requires the
examiner to strap the patient’s limbs to some kind of cinematic device.
3.3.4 Dyskinesia Quantification
Figure 3-9: Signals of a triple-axis gyroscope and its power spectrums during an arms-extended task. This
figure describes how the features change as a patient experienced one cycle of levodopa-induced dyskine-
sia. Tremor was apparent in hours 1 and 3, while the dyskinesia was experienced in hour 2 (Based on
Mera et al., 2013).
The research conducted by Burkhard et al. (1999) shows that the RMS of the power spectrum
Time domain Frequency domain Clinical scores
0 2
2 0
0 1
Dyskinesia Tremor
(s)
Hour 2 Hour 2
2
State of the Art
19
of hand-attached gyroscope signals, from 0.25 to 3.25 Hz, during dyskinesia task correlated
well with the five-point clinical ratings for dyskinesia severity.
As Figure 3-9 shows, Mera et al. (2012) revealed that the RMS of the frequency band, from 0
to 3Hz, from all the three gyroscope channels correlates well with the modified Abnormal
Involuntary Rating Scale (m-AIMS). Two stationary motor tasks were performed during the
arms-extended task: arms resting and arms extended. The dyskinesia severity was calculated
based on the inertial sensor signal’s range from 0 to 3 Hz. However, the algorithm could not
be applied to voluntary motor tasks.
3.4 Commercial Systems for Quantification of Neurological Symptoms
Kinesia™ (CleveMed Inc., USA) is a compact, clinical device that is used to objectively
quantify the motor symptoms of movement disorders such as PD and ET.
Figure 3-10: Kinesia™ (old version) for tremor and bradykinesia assessments (© CleveMed, U.S.A, 2010).
On the left side of the figure, a subject performed the assessment task according to the video instruction.
The right side of the figure shows the hardware of a Kinesia™ system.
a) b) c)
Figure 3-11: Kinesia™ (new version) for tremor and bradykinesia assessments (© Great Lakes
NeuroTechnologies, U.S.A, 2013). a) hardware and GUI; b) tremor tuning map for a patient during an
outpatient programming session; c) bradykinesia tuning map for a patient during an outpatient pro-
gramming session. The voltage is the stimulation intensity of the electrode. The color in the tuning maps
means the severity of the symptoms from red (most severe) to green (no symptoms). CleveMed Inc. spun
Great Lakes NeuroTechnologies Inc. in 2011, which focuses on the assessment of motor symptoms based
on inertial sensors.
As shown in Figure 3-10, a Kinesia™ was worn on the finger and wrist of the patient while
symptom information was wirelessly transmitted to a nearby computer for display, analysis,
automated symptom severity scoring, report generation, and storage. Three orthogonal gyro-
State of the Art
20
scopes and three orthogonal accelerometers were placed on the finger, to capture motion with
six degrees of freedom (DOF). This device can be used by PD patients to monitor the kine-
matics of motor symptoms such as tremor and bradykinesia. Several time- and frequency-
based parameters are computed for each kinematic channel (axis) including peak power, fre-
quency of the peak power, root mean square (RMS) of the angular velocity, and RMS of the
angle. This device has been approved by the FDA (Giuffrida et al., 2009). KinetiSense, an
upgraded version of Kinesia™, is a small, lightweight, wireless device that integrates motion
detection and EMG. Three orthogonal accelerometers and gyroscopes provide 3D motion
tracking while two EMG channels record muscle activity.
As shown in Figure 3-11, Great Lakes NeuroTechnologies Inc. released the new generation of
Kinesia™ in 2013, which can generate functional motor symptom response tuning maps. The
wireless ring sensor communicates with the tablet computer via a Bluetooth interface. Tuning
maps can be used to monitor the stimulation setting after DBS surgery for a particular symp-
tom or averaged across multiple symptoms. Its scoring algorithms are clinically validated.
HandTutor™ (MediTouch Inc., Israel) is a state-of-the-art glove and software that enables
intensive clinic- or home-based hand rehabilitation (quantitative assessment and customized
training). As shown in Figure 3-12, it evaluates the extension/flexion of the fingers and wrist.
It comprises six triple-axis accelerometers, five on the fingers and one on the wrist. The data
recorded by the device include motion locus, spectrum, maximum velocity, and ROM (range
of motion) (Carmeli et al., 2009).
Figure 3-12: HandTutor™. Electro-optical sensors, bend sensors, and accelerometers are placed in the
upper side of each finger. The training parameters of each finger are displayed in the GUI. It is power
supplied by the USB interface of a computer (© MediTouch, 2010).
The Motus Movement Monitor (MOTUS Bioengineering Inc., USA) can be used to monitor
the patient’s responses for the electrode implantation during DBS surgery. As shown in Fig-
ure 3-13, only one gyroscope is used on the surface of the palm. The Motus system assists the
neurologist in the selection process by providing quantification of movements, clearly docu-
menting tremor characteristics, the extent of bradykinesia (via pronation/supination move-
ments), and the prominent side of the disorder.
State of the Art
21
Figure 3-13: Motus Movement Monitor. The raw data, Fast Fourier Transform (FFT) chart and parame-
ters are displayed in the GUI (© MOTUS, 2010).
In addition, glove-based hand motion tracking systems represent an appropriate method aimed
at acquiring hand movement data. There are many products currently available on the market
that can achieve these or similar goals. These include the Cyberglove, P5 Glove™, 5DT
Dataglove, Acceleglove, Hand Mentor, and HandTutor. They allow continuous tracking of
hand joint angles and linear displacements.
3.5 Research Systems for Quantification of Neurological Symptoms
Su et al. (2003) of the University of East Anglia, UK, developed a three-dimensional motion
recording system (data-gloves) for PD assessment. As shown in Figure 3-14 (a), this system
uses 11 electromagnetic sensors on the hand. Time traces, frequency analysis, and speed
analysis of these time traces are the key parameters. Tremor parameters, the rigidity of the
wrist during rolling movements, the dexterity of finger pinching, and hand-gripping move-
ments are recorded. As shown in Figure 3-14 (b), they later used EMG together with the pre-
vious motion system to obtain better results (Su et al., 2003, 2007).
a) b) c)
Figure 3-14: 3D motion recording system. a) sensor layout; b) simultaneous recording system; c) user
graphical interface of the simultaneous data recording (Taken from Su et al., 2007).
Arash of the EPFL1, Switzerland, designed a new measurement system consisting of five in-
1 Ecole Polytechnique Fédérale de Lausanne
State of the Art
22
dependent, lightweight, autonomous sensing units, which were based on gyroscopes that can
continuously record body movements during daily life. An accurate algorithm based on spec-
tral estimation was proposed to detect and quantify tremor during the daily activities of PD
patients with a resolution down to three seconds using gyroscopes attached to the forearms.
This system, which is shown in Figure 3-15, also successfully detected ON and OFF periods
in PD patients under ambulatory conditions (Arash et al., 2006).
a) b)
Figure 3-15: Ambulatory system for quantification of tremor and bradykinesia. a) the sensitive axes of the
3D gyroscope; b) a closer photo of the module (Taken from Salarian et al., 2006).
Bamberg from MIT (Massachusetts Institute of Technology), USA, built a shoe-integrated
sensor system for wireless gait analysis. She used gyroscopes, accelerometers, and force sen-
sors for the calculation of gait parameters. Detailed information on the gait cycle was ob-
tained. Her PhD thesis described the signal processing methods of gyroscopes and acceler-
ometers in great detail (Bamberg et al., 2008).
Funded by EU Framework 7, and coordinated by Prof. José Luis Pons of the Bioengineering
Group, CSIC (Spanish National Research Council), Spain, the tremor project team is develop-
ing an ambulatory brain-computer-interface-driven tremor suppression system based on func-
tional electrical stimulation. This system will be used to detect and monitor tremor through a
multimodal brain-computer-interface (BCI). The proposed BCI method will combine CNS
signals (Electroencephalography, EEG) and peripheral nervous system signals (Electromyog-
raphy, EMG) data with biomechanical data (inertial sensors) in a sensor fusion approach. It
will model and track both tremor and voluntary motions (Ibánez et al., 2010).
Figure 3-16: Validation of elastic stiffness (K) calculations using a model arm. The quantification device
was tested on a prosthetic limb to which combinations of elastic cords were attached to produce different
levels of constant stiffness. An examiner flexes and extends the subject’s joint. A gyroscope in the arm was
used to measure the elbow angular displacement, while a differential force transducer attached to the
wrist was used to monitor the amount of force employed (Based on Patrick et al., 2001).
Load cell and LVDT (linear variable displacement transducer):
Linear regression-calculated K (N·m/degree)
Validation of elastic stiffness (K) calculations
Quantification d
evic
e K
(N·m
/deg
ree)
State of the Art
23
Patrick et al. (2001), from the University of Alberta, Canada, have developed a stiffness quan-
tification device. As shown in Figure 3-16, the device is based on two air-filled pads held dis-
tal to the joint and a gyroscope mounted on one of the force pads. Both pads are connected to
a differential force transducer. Measurements of mechanical impedance (the magnitude of the
vectorial sum of elastic stiffness and viscous stiffness) corresponded well with the clinical
ratings of parkinsonian rigidity.
Other researchers have designed mechanical devices to simultaneously measure the torque
and angular position of the elbow or wrist joint during flexion-extension movement (Shapiro
et al., 2007).
Sepehri et al. (2007), from the Islamic Azad University-Mashhad Branch, Iran, presented a
test rig to measure the range of motion, and viscous and elastic components of elbow stiff-
ness. As Figure 3-17 shows, the subject’s elbow joint was fixed in the mechanical test rig. A
balanced strain gage force transducer and a 10 K potentiometer were used to measure at-
tached force and angular displacement, respectively. Their results revealed that the elastic and
viscous components of mechanical impedance measured with the device correlated well with
the UPDRS ratings.
Figure 3-17: Mechanical rig test system and its operation (Taken from Sepehri et al., 2007). The elbow
movement was driven by an examiner.
As Figure 3-18 shows, Park et al. (2011), from the Korea University, analyzed the wrist’s
viscoelastic properties in PD patients. Their study suggested the mean viscosity during both
flexion and extension correlated best with the clinical rigidity score.
Figure 3-18: System structure and experimental setup for the analysis of the wrist joint’s visco-elastic
properties (Based on Park et al., 2010). The wrist movement was controlled by a motor.
Handle
Load cell
Hand plate
Half-circle plate
Lower arm plate
Potentiometer
Cotton-buffered splint
Handle
Load cell
Accelerometer
Potentiometer
State of the Art
24
As shown in Figure 3-19, Niazmand et al. (2011a), from the TU Muenchen, Germany, pre-
sented a smart glove for quantitative evaluation of PD symptoms. Three primary symptoms
could have been assessed. However, there was no correlation between the clinical rigidity
severity and force sensor value. In addition, there was no discussion of tremor amplitude. The
bradykinesia assessment task was finger taps.
Figure 3-19: Smart glove for quantitative evaluation of PD symptoms (Taken from Niazmand et al.,
2011a). 1) force sensor; 2) command module; 3) sensor board; 4) touch sensor (positive connector); 5)
touch sensor (ground connector). The force sensor was attached to the surface of the command module.
Signal Processing Methods
In these products and research projects for the quantification of tremor or bradykinesia severi-
ty, a fast Fourier transform (FFT) or power spectral density (PSD) calculation in real-time is
carried out. Fuzzy analysis, a machine-learning algorithm, and other algorithms are used for
frequency and amplitude analysis. Regression analysis is used in most rigidity assessment
systems.
3.6 Inertial Sensors and Sensor Fusion for Motion Tracking
3.6.1 Inertial Sensors and Inertial Measurement Unit
Some types of motion sensors have been commercially available for several decades in appli-
cations for ships, aircraft, and automobiles. However, these sensors’ characteristics, such as
dimension, power consumption, and price, have prevented their wide utilization in consumer
electronic devices up until the past few years (Shaeffer, 2013).
Accelerometer (G-sensors), gyroscope, magnetic sensor (E-compass), and pressure sensors
(barometers) are the four fundamental motion sensors for mobile device and other consumer
electronics.
Accelerometers measure linear acceleration (dynamic acceleration) and tilt angle (static ac-
celeration) with limited motion sensing functionality. Gyroscopes measure the angular veloci-
ty in one or more axes with high signal-to-noise ratio (SNR). Gyroscopes can accurately
measure complex rotational motions in free space. Another difference between gyroscopes
and accelerometers/compasses is that gyroscopes function fairly autonomously, other than
depending on any external forces such as magnetic fields or gravity. Compasses detect only
the heading of a triple-axis space based on the earth’s magnetic field. Pressure sensors realize
relative and absolute altitude by sensing the relationship analysis between atmospheric pres-
sure and altitude.
Using a gyroscope or an accelerometer as the single source for angle calculation, also shows
disadvantages (Patel et al., 2008; Okuno et al., 2009). The output of a gyroscope is read at
State of the Art
25
certain time intervals. Thus, the information between these periods is missed. In order to im-
prove the accuracy of angular displacement, more gyroscope samples are needed, which takes
more processing time. Due to the inaccuracy of each gyroscope reading, the angular dis-
placement calculated will drift over time. The angle calculation with accelerometer data is
based on gravity. The accelerometer will give an accurate reading of tilt angle in a static state.
But accelerometers are slower to respond than gyroscopes and are prone to vibration or noise.
An accelerometer can be used to correct gyroscope drift errors and is more sensitive to non-
rotation movement.
Another type of sensor commonly used to reduce drift is the magnetometer, which measures
magnetic field strength in a given direction. The magnetometers measure the strength and
direction of the local magnetic field, allowing the north direction to be found. The yaw an-
gle is calculated from the magnetic field.
The silicon MEMS-based technology reduces the cost and package size of an inertial sensor.
At present, most analog inertial sensors are replaced by digital inertial sensors, which means
that there is no analog-to-digital converter (ADC) outside the sensor anymore. In addition,
three axes sensors are combined in a single chip. Programmable filter and full-scale range
setting are also available inside the chip. All these features give MEMS-based sensors better
noise performance than before.
Motion processing with MEMS technology, which measures and intelligently processes the
movements of subjects in three dimensional spaces, is the next major revolutionary technol-
ogy that will drive innovation in mobile device design, human-machine interface design, and
applications for navigation and control. Consumer-grade IMUs based on MEMS provide a
simpler user interface for intuitionistic navigation and control of handheld mobile devices.
Due to the advantages of IMUs, the operational complexities that have confused many owners
of sophisticated consumer electronic devices can be resolved (Shaeffer, 2013).
Over the past 30 years, inertial sensors have been used in major automotive and industrial
markets. With the development of mobile devices, especially the revolutionary iPhone and
iPad from Apple Inc., U.S.A, motion MEMS sensors with low-cost, ultra-compact, low power
consumption and multiple-axis sensing have been developed in last five years. ST,
Invensense, Bosch, Freescale, and Kionix are the dominant motion MEMS manufacturers at
present.
InvenSense claimed it was the first company to develop an integrated triple-axis MEMS gyro-
scope and six-axis motion-tracking device, with digital-output, for consumer electronics ap-
plications. In February 2011, Invensense Inc. launched MPU-6000, which integrates three
accelerometers and three gyroscopes into a single package (4mm 4mm 0.9mm). In August
2011, nine-axis motion-fusion algorithms (together with a triple-axis compass outside con-
nected) and an upgraded version (MPU6050) were presented.
3.6.2 Accelerometer-Magnetometer and Attitude Heading Reference System
A six-axis accelerometer-magnetometer unit, which contains a triple accelerometer and a tri-
ple magnetic sensor with about 0.6 mA of power, accurately determines heading and orienta-
tion. Some companies regard this type unit as a simulated IMU. However, this unit can only
measure slow changes in a three dimensional space.
State of the Art
26
Gyroscopes and accelerometers are great, but they cannot provide precise and accurate calcu-
lations such as the absolute heading value. An attitude heading reference system (AHRS) con-
sists of a triple-axis gyroscope, a triple-axis accelerometer and a triple-axis magnetometer
(compass). The use of the compass, which provides a heading reading, offers enhanced angu-
lar position accuracy, and reduces gyroscope drift.
Raw output from multiple discrete sensors requires development and incorporation of a com-
plex set of sensor fusion algorithms, calibration firmware, and performance testing prior to
use (Paces & Popelka, 2012).
In August 2011, nine-axis motion-fusion algorithms (together with a triple-axis compass out-
side connected) and an upgraded version (MPU9050) of the Invensense MPU6050 were pre-
sented. After that, many manufacturers were forced to incorporate discrete motion sensor
components to deliver nine-axis Motion Interface functionality.
Sensor fusion is not limited to a nine-DOF (degree of freedom) solution. For the requirement
of indoor navigation, the 10-DOF or 10-ASF (Acclaim Skeleton File) solution includes a tri-
ple-axis accelerometer, triple-axis gyroscope, triple-axis magnetometer, and a single-axis ba-
rometer. Adding a barometer enables altitude detection, since pressure changes with altitude
at a rate of about 10 Pa/m. At the moment, some mobile motion devices or some functions of
other consumer electronics, such as iWatch, Google Glass, are based on nine-DOF or 10-DOF
sensor fusion realization.
3.6.3 Sensor Fusion Algorithms
The primary feature of IMU and AHRS is the sensor fusion. For the rotation movement
measurement with Euler angles, complex finite impulse response (FIR) or infinite impulse
response (IIR) filters such as Kalman filters, Parks-McClellan filters, are widely used. The
Direction Cosine Matrix (DCM) algorithm is used to calculate the orientation of a rigid sub-
ject, offering another way to construct a rotation matrix (Edwan et al., 2011).
In addition, most inertial manufactures present unique on-board sensor fusion algorithms. For
example, the inertial sensors from Invensens Inc. include an in-chip sensor fusion module
named Digital Motion Processor™ (DMP™), which is capable of processing the complex
nine-axis MotionFusion algorithms. ST Microelectronics Inc. presented sensor-fusion soft-
ware named iNEMO Engine, which is based on dedicated filtering and prediction algorithms.
DMP™ and iNEMO Engine combine different data from multiple sensors. These sensors can
directly provide a series of outputs such as rotation, linear acceleration, gravity, and quater-
nion. The control of these sensors can be performed using an eight-bit microcontroller (MCU)
and are independent of environmental conditions to achieve the best performance.
In addition, there are some inertial sensors embedded into a single chip with an MCU and
other function modules such as a radio frequency (RF) module.
3.6.4 Discussion
The key difference between an IMU and an AHRS is that the AHRS provides accurate atti-
tude and heading solutions (yaw angle) whereas the IMU only delivers the absolute attitude
solution (pitch and roll angles). As hand tremor is mainly a rotational movement, an AHRS is
good for measuring hand tremor. However, the magnetometer is susceptible to ferromagnetic
material, and thus needs accurate soft and hard iron calibration. As the period of hand tremor
State of the Art
27
movement assessment is short, the drift of yaw is slow and can be removed with a threshold
setting for gyroscope data. An IMU can also achieve good results when measuring hand
tremor.
The accelerometer is good at measuring linear motion and the gyroscope is good at measuring
rotational movement. When a gyroscope and an accelerometer are combined, a better result
can be realized. The combination of small, low-cost but high performance triple-axis gyro-
scopes, which have recently become available, and the existing triple-axis MEMS acceler-
ometers, enables the possibility of this six-axis measurement and control. Mobile devices
have already embraced the novel features provided by the MEMS IMU, which combines a
triple-axis gyroscope and a triple-axis accelerometer. The six-axis motion processing provides
the mobile device’s absolute position in a three dimensional space with greater accuracy, pre-
cision, and responsiveness.
3.7 Limitations of Existing Technology
The disadvantages of the state of the art are discussed in this section.
3.7.1 Real and Potentially Solvable Limitations
Subjective assessment by the surgeons and MER system are widely used to support symptom
assessments during DBS surgery. The two intraoperative approaches for neurological symp-
tom assessments show disadvantages as follows:
a) Subjective assessment by surgeons according to the five-point clinical ratings:
The judgments of the surgeons are based on their experience and differ from each oth-
er.
UPDRS and TETRAS are discrete and subjective ratings. They require a neurologist to
visually assess the patient based on experiences, and the neurologist cannot capture com-
plex symptom variations that happen in response to the stimulations during DBS surgery.
The symptom scores for the same patient may differ widely depending on the examiner
(Jankovic et al., 2007).
The coarse resolution of the ratings is insufficient for assessing small changes in tremor
severity. Furthermore, the extent of inter-clinician and inter-subject rating variability is
unknown (Machado et al., 2003).
b) Micro-electrode recording (MER):
It is an indirect motion tracking. The hair-thin microelectrodes are placed within the
intended target to record brain cell activity. The surgeon needs to inspect and listen to
the pattern of the cell activity. This physiological confirmation is also based on the
experience of the surgeon (Winestone et al., 2012).
The MER signals are spike signals, which are not good for signal processing
(Winestone et al., 2012).
Research shows that only about 67% of the cases initially planned with MER method
were taken for the final target (Bour et al., 2010).
State of the Art
28
According to the research of Bour et al. (2010), the location of the best MER activity did
not necessarily correlate with the position that produced the optimal clinical response to
microelectrode testing intraoperatively.
EMG does not directly measure body movements and a large number of electrodes may be
needed to investigate complex movements. Same with MER, the information collected on
frequency is good, but that on the magnitude is less reliable. Contact resistance is also a sig-
nificant variable. Furthermore, no information on displacement, velocity or power can be ob-
tained. The EMG waveform of tremor is spiky and exhibits an impulse chain in the morphol-
ogy. Impulse-like data are hard to analyze with Fourier spectral methods or other traditional
amplitude methods. The measured amplitude is variable and thus the result is inaccurate
(Saara et al., 2007).
Devices other than the MER described above are used for general purposes, and are not spe-
cifically designed to be used as guiding tools during DBS surgery and do not meet the needs
of the operating room. They have demonstrated limited usability in clinical settings due to
deficiencies in wearability, fidelity, and flexibility. The severities of ET and parkinsonian
symptoms, which are the effect of DBS therapy, require real-time and accurate assessment of
major parameters according to the UPDRS. The severity changes of the primary symptoms
are crucial when evaluating the effect of DBS. Easy manipulation and comfort are also im-
portant. The graphical user interfaces (GUI) of these systems are also unsuitable for the DBS
monitoring.
Some of the glove-based systems, such as Cyberglove, are accurate but relatively costly
($10,000 per Cyberglove). Others are more affordable, for example, the P5 Glove™ (ap-
proximately $100 per glove) (Dipietro et al., 2008). Most use piezoresistive sensors, fiber
optic sensors, and Hall-effect sensors, rather than inertial sensors. However, a major limita-
tion of these systems is their limited portability caused by the presence of cloth support. The
cloth support was believed to affect the measurement performance (Dipietro et al., 2008). The
data obtained from these gloves have to be modified for further processing. In addition, the
settings of the sensors are not easy to operate and calibrations are needed for new users. A
new data glove with gyroscopes and accelerometers is more feasible.
For bradykinesia assessment, finger tapping, and pronation/supination movement are not easy
to perform during DBS surgery. Rigidity is clinically defined as increased resistance to pas-
sive movement of a joint. For the joint movement, derived by a motor, has bigger dimensions
and should be fixed on a table (Patrick et al., 2001).
The displacement measurement for hand rigidity and hand grasping can be estimated through
double integration of raw acceleration data over time. However, gravity vector and bias
should be deducted from the raw acceleration. Then IMUs, which includes three-axis gyro-
scopes, should be utilized in this study.
3.7.2 Limitations That Cannot Be Solved at This Time with Reasonable Effort
The basic changes in the EMG and MER signals, which are caused by PD or ET, are an in-
creased tonic background activity and an alternating pattern of signal bursts. During the EMG
or MER analysis of the symptoms of PD and ET, special attention has been paid to the analy-
sis of these bursts by measuring their counts, magnitudes, durations, and frequencies
(Rissanen et al., 2007). For the measurement of vigorous movement, an IMU provides good
State of the Art
29
results. However, for slight motion disorders, the difference between EMG and IMU requires
further validation.
Tremor, bradykinesia, and limb rigidity are the characteristic features of PD. However,
bradykinesia responds to DBS after a period of hours, whereas tremor and rigidity respond
within seconds (Prodoehl et al., 2007). For the design of the assessment tasks, there is no con-
cern about this situation yet.
The UPDRS is a subjective rating and the rigidity scores for the same patient may differ wide-
ly depending on the examiner. Because of the role of the patient’s passive movement in as-
sessing rigidity, the performance of the examiner also affects the assessment results through a
rigidity assessment system without a motor (Patrick et al., 2001).
Rigidity occurring in PD patients commonly has a “cogwheel” character, which is not repre-
sented by the UPDRS (Van Dillen et al., 1988).
The extent of hand tremors in PD varies between patients. The ratio of rotational motion to
linear motion is not fixed. An optimal ratio that fits all patients can only be obtained after a
large number of clinical measurements have been taken (Timmer et al., 1993).
The goal of DBS surgery is to obtain the best treatment strategy, because PD affects every
patient differently. Tremors and other symptoms do not appear at the same time for every
patient. The assessment results of a patient with the designed system also depend on the skill
level or psychological factors of the patient. The significant individualized factors of the pa-
tient include the age, years of disease, level of functional impairment, concurrent medical
issues, and sensitivity to medications (Spieker et al., 1995). Therefore, it is important to con-
sider various individualized factors of the patient when utilizing assessment results. These
factors can be investigated only after long-term use of the glove monitoring system. Objective
assessment methods which fit for all patients with PD or ET are needed in the assessment of
their symptoms.
Glove Monitoring System
30
4. Glove Monitoring System
Because all the primary symptoms of PD and ET can be assessed on the hand, a glove moni-
toring system based on inertial sensors and force sensors is supposed to monitor the severities
of tremor syndromes, bradykinesia, and rigidity during DBS surgery.
Chapter 4.1 introduces the task description of the glove monitoring system. Several assess-
ment tasks for tremor syndromes, bradykinesia, and rigidity were chosen for the glove moni-
toring system according to the requirements of DBS surgery. The supposed parameters and
their accuracy settings, together with the other requirements, are presented in this section too.
The expected advantages of the glove monitoring system are described in Chapter 4.2.
4.1 Task Description
Currently there is no system available for assessing all the primary PD symptoms. Therefore,
it is important to realize three assessment tasks in one system. The goal of this project is to
develop a measuring system to be used in DBS surgery, which quantifies the severity of hand
tremors, bradykinesia, and rigidity.
4.1.1 Assessment Movements in This Study
According to the literature and the requirements of neurosurgeons, several assessment tasks
should be chosen to assess the severity of the symptoms of PD and ET. All the assessment
tasks are supposed to have the same duration (from five seconds to one minute). The surgeon
instructs the patient to perform different tasks. The observers can view all the parameters and
raw signals on the computer.
According to the MDS-UPDRS, tremor assessments include three tasks to test for the rest
tremor, postural tremor, and action tremor.
According to the requirement of surgeons, whole-hand grasping was chosen as the
bradykinesia assessment task because the patients can perform it easily during DBS surgery.
Figure 4-1: Rigidity assessment task. 1) subject; 2) rigidity cuff; 3) examiner. Here l is the arm length of
the patient. The examiner flexes and stretches the elbow through the rigidity assessment cuff attached to
the wrist. Several cycles are performed during the 10-second assessment task. The examiner should make
sure the elbow position of the patient is stable (Based on Dai et al., 2013b).
Passive flexion and extension of the elbow was used to assess rigidity. Figure 4-1 shows the
rigidity cuff and rigidity quantification task in this project. The rigidity cuff is strapped to the
Glove Monitoring System
31
distal end of the patient’s forearm. An examiner flexes and extends the patient’s elbow joint
through force at the point of the rigidity cuff on the wrist.
After discussion with the surgeons, the duration of a single task in this study was set at ten
seconds. Each time two or more measurements for a single task should be repeatedly per-
formed for better results.
4.1.2 System Objectives and Parameters
The parameters that should be obtained from both time- and frequency-domain signal proc-
essing methods and displayed in the GUI are listed as follows:
Severity of hand tremors (frequency and amplitude of tremors).
Severity of finger bradykinesia (mean and standard deviation values of angular dis-
placements in hand grasps; dominant frequency in hand grasps).
Severity of elbow rigidity (viscosity, elasticity, and the dominant frequency of elbow
movement).
The angular displacements in the bradykinesia task represent peak-to-peak values of the hand
grasping ranges during one time bradykinesia task.
In addition, the power spectral density of the signals from the gyroscope and accelerometer
(from 0.25 Hz to 12 Hz), together with the raw data, needs to be displayed in real-time.
The severities of tremors, bradykinesia, and rigidity should be displayed at levels from 0 to 4
according to the UPDRS ratings. There is, however, no literature about the bradykinesia pa-
rameters of hand grasping movement and UPDRS scores. The severity of parkinsonian rigid-
ity is very difficult to assess. Therefore, the severity scores of bradykinesia and rigidity ac-
cording to the present parameters need to be further investigated.
Fluctuations in the motor performance (ON/OFF fluctuations) of PD patients can also be indi-
cated by the changes in these parameters.
For DBS electrode positioning, only the tremor amplitude is required, because the tremor fre-
quency has no relation to the symptom severity of parkinsonian tremor or ET. The tremor
amplitudes, which are regressed from the peak powers of the IMU signals, can be gauged as
the clinician ratings. However, the parameters of bradykinesia and rigidity cannot be normal-
ized directly as UPDRS ratings at present.
Unlike for tremor, peak power during hand grasping is not correlated with the clinical
UPDRS score. Instead, the mean value and standard deviation (SD) of hand grasping ranges (
and ), where is the peak-to-peak values of the hand grasping cycles, can be used as
the parameters of bradykinesia. However, the correlation between these two values and the
UPDRS score needs further study (Post et al., 2005).
In the clinic, rigidity is assessed by a neurologist who moves the subject’s limb, scoring the
result according to the UPDRS ratings. However, rigidity scores for an individual patient may
vary depending on the examiner. Such scales are susceptible to the problems of sensitivity
and reliability. Mechanical impedance, which means the magnitude of the vector sum of elas-
Glove Monitoring System
32
tic stiffness and viscous stiffness, is nonlinearly related to UPDRS rigidity ratings (Charles,
2003).
4.1.3 Accuracy Settings of the Parameters
After the clinical measurements and experiments, the correlations between the measured pa-
rameters (dominant frequency and amplitude in each tremor task; dominant frequency, mean,
and standard deviation values of grasping ranges in bradykinesia task; viscosity, elasticity,
and frequency of elbow movement) and the UPDRS or TETRAS scores can be determined.
For calculating the accuracy settings in Table 4-1, the parameters obtained from the glove
monitoring systems with the IMU and force sensors must be compared to the judgments of
doctors according to the UPDRS ratings. The parameters should meet the accuracy setting as
shown in Table 4-1. In Table 4-1, is the coefficient of determination and RMSE is the root-
mean-square-error (RMSE).
Table 4-1: Parameters and their required accuracies. For the RMSE in rigidity parameter (mechanical
impedance), there is no relative literature.
Task Parameter Unit,
[Range]
Required accuracy Reference
2r RMSE
Rest tremor R1 1, [0–4] 0.85 0.32 Giuffrida et al.,
2009
Postural tremor R2 1, [0–4] 0.88 0.32 Giuffrida et al.,
2009
Action tremor R3 1, [0–4] 0.60 0.45 Giuffrida et al.,
2009
Bradykinesia °, [0–360] 0.72 0.52 Heldman et al.,
2011
Bradykinesia °, [0–360] 0.62 0.65 Heldman et al.,
2011
Rigidity Z N/°, [0–50] 0.35 - Patrick et al.,
2001
With patients and volunteers, the functionality and accuracy of the glove monitoring system
should be tested and verified.
4.1.4 Requirements
General requirements of the glove monitoring system are listed as follows:
Easy to perform for both the surgeons and patients (based on the intraoperative test
tasks).
Help the surgeons to quantify symptom severities during DBS surgery and as feedback
to the electrode stimulation.
Accurate and reliable: the parameters should meet the accuracy settings in Table 4-1.
Quantify symptom severities according to the UPDRS or TETRAS ratings.
Provide parameter-recording lists to compare symptom severities in all tested elec-
trode positions.
Glove Monitoring System
33
High security and low risk.
Does not restrict the patient’s movement (comfortable).
Portable and with small dimensions.
Meet the requirements of the operating room.
Some technical requirements are listed as follows:
Accurate calibration plays a key role in the IMU measurements: the accuracy of the
force measurement and the angular displacement measurement should be higher than
20% and 10% respectively.
A large amount of clinical measurements is needed for the modification of algorithms,
especially the coefficients of the regression models used in this project: 30 patients for
each symptom should be measured with this glove monitoring system.
The raw data from the sensors should be displayed to show the original hand move-
ment. The raw data and measured parameters should also be stored in a file. A real-
time PSD display of the inertial sensor signals over the entire frequency range (0.25–
10 Hz) should be incorporated.
The user graphical interface should be easy to use and intuitive, both for technicians
and surgeons.
A database with customized reporting functions and data management is required.
The assessed parameters should be updated quickly when the assessed symptom
changes. These parameters should be saved and displayed for contrast when the elec-
trode is moved to different positions.
To avoid radio frequency interference (RFI) in the operating room, a wired measure-
ment device is required during DBS surgery. A wireless measurement device is pre-
sented only for the measurement outside the operation room.
4.2 Expected Advantages
The major advantages of this glove monitoring system are:
Objective assessment, thus easing the surgeons’ workload.
This project focuses on DBS, specifically integrating the overall measurement of finger and
elbow movement. By incorporating a glove for measuring and displaying the parameters and
waveforms on a computer, the system will be easy to use and relieve the surgeon. The PSD
method in this project does not simply divide the time dimension into windows (3–5 seconds
or longer), and a single data point can be used more than once. Thus the PSD chart is continu-
ous with the sampling interval. The single-sided, scaled, auto-power spectrum of the time-
domain signals will be displayed together with the raw signals. This feature will allow better
detection of changes of the symptoms.
Three primary symptoms of PD and ET are implemented in a portable system.
Glove Monitoring System
34
Unlike previous studies, the three primary symptoms are included in a system. In addition,
this system meets the requirements of DBS surgery.
Electronics safety because the electronic parts have no contact with the patient’s
body.
These sensors have no direct contact with the human body, thus this system is safer than
EMG and MER.
MEMS IMU technology makes the system smaller but with a higher performance
compared to previous motor assessment systems.
With ever-smaller dimensions and higher performance, it is easy to detect motor disorders in
PD and ET. The latest motion MEMS sensors make it is possible to realize motor disorders
with small dimension and higher performance.
Higher resolution compared to the UPDRS or TETRAS ratings: 0.01.
Each sub-scale of the UPDRS are from 0 to 4, where 0=normal, 1=slight, 2=mild,
3=moderate, and 4=severe. Most of the previous research only compared the outputs of their
systems to this five-point rating scale (Elble et al., 2006). During DBS, a higher resolution for
all parameters is needed.
Accelerometer data showed strong correlations with UPDRS rest tremor scores for both trem-
or duration and amplitude, especially when the hand tremor involved non-rotational move-
ment. However, accelerometers measure linear acceleration and are influenced by gravity,
whereas gyroscopes measure gravity-independent angular velocity. Tremor measured using
gyroscopes correlated well with UPRDS scores, with higher sensitivity and specificity than
previous studies that used accelerometers (Giuffrida et al., 2009). By using a six-axis sensor
motion fusion method, the measured parameters are more sensitive to hand movements in all
directions.
After surgery, this glove monitoring system also can be used to track the progress of the pa-
tients and to provide feedback to physicians on the long-term results of surgery. When the
patient returns for periodic readjustment of the neurostimulator, the glove monitoring system
can be used to provide quantitative measurements of tremors and bradykinesia to assist in
selecting the optimum parameters for electrode stimulation.
System Concept
35
5. System Concept
Two designs using wireless communication interfaces are first introduced in Chapter 5.1.
These designs could be used for a series of tremor and bradykinesia assessment tasks. All
fingers were attached with sensors. These designs are not for applications outside of the oper-
ation room, other than for DBS surgery.
According to the requirements during DBS surgery, the static and dynamic system descrip-
tions of the glove monitoring system are described in Chapter 5.2 and Chapter 5.3 respective-
ly. The first step in designing the glove monitoring system includes two parts: one for tremor
and bradykinesia assessment, the other for the rigidity assessment. In the end, a combined
version based on the two separate systems is introduced.
5.1 Designs with Wireless Communication Interfaces
At the beginning of the study, there were two designs for PD assessment based on wireless
communication, touch sensors, and inertial sensors (gyroscope and accelerometer). Several
prototypes based on these designs have been carried out but they were not for the purpose of
DBS. They could be used for the parameter settings of the neurostimulator after DBS surgery.
According to the literature, accelerometers and gyroscopes placed on each finger can collect
enough information about parkinsonian tremors (Dipietro et al., 2008). A gyroscope is better
for bradykinesia assessment. This suggests that a triple-axis accelerometer and a triple-axis
gyroscope on each finger will provide sufficient information regarding the hand activity of
patients with PD or ET. Compared to a gyroscope and an accelerometer on one finger, the use
of six gyroscopes and six accelerometers on all fingers and the wrist would provide more in-
formation about hand movement. Thus, the quantification of tremors and bradykinesia would
be more objective.
Figure 5-1 shows the diagram of the first tremor and bradykinesia assessment system with a
wireless interface. In addition to inertial sensors, four touch sensors were attached to the fin-
gers as well. The touch sensors were used for bradykinesia assessment (finger tapping task).
Figure 5-1: Components of the tremor and bradykinesia measuring system with wireless communications.
The mean tremor amplitude in all fingers, distance, and duration between two finger taps were
calculated as the parameters of tremor and bradykinesia, respectively. As a result, from the
measurement of the difference between finger motions, the obtained parameters could be
more accurate.
Sensor nodes (IMUs and touch sensor positive points)
Command module with wireless interface
Wireless receiver
Sensor node (IMU and touch sensor ground point)
System Concept
36
After the analysis of these parameters, fewer sensors were adopted but with the same quality.
For a simplified version of this design, the thumb and forefinger should be placed with the
inertial sensors and touch sensors.
During DBS surgery, the glove monitoring system is wired to a computer, whereas after sur-
gery a wireless system can be used. This system was for the assessment outside of the opera-
tion room and could not assess rigidity severity.
5.2 Static System Description
Figure 5-2 shows the original system diagram of the glove monitoring system. The glove
monitoring system is based on MEMS IMUs, FSRs, and a medical textile glove. A sensor
board with an IMU placed in the upper side of a finger is used to assess tremors and
bradykinesia. The rigidity assessment part consists of a rigidity cuff, which was able to be
attached to the wrist and includes two differential force sensor boxes and another IMU. The
rigidity assessment cuff is designed to attain the model of a joint’s movement state (angular
displacement and velocity) and its measured torque (N•m), which includes non-neural torque
and neural torque. The sensor data is acquired by a command module and sent to a computer
via a USB cable. The sensor board, command module, and rigidity cuff can be integrated into
a textile glove. Further signal processing is carried out on a computer using MATLAB, Visual
Studio, or LabVIEW.
a) b)
Figure 5-2: a) general system diagram of the glove monitoring system; b) rigidity assessment cuff. The
components are: 1) sensor board (IMU); 2) textile glove; 3) command module; 4) rigidity assessment cuff;
5) USB cable; 6) graphical user interface (Based on Dai et al., 2013a).
The computer communicates with the command module via a USB cable (Lorenzo, et al.,
2011). The wired communication, instead of wireless communication, has the advantage that
this system even can be used in the operation room.
A 6-axis IMU module (combines a triple-axis gyroscope and a triple-axis accelerometer) on
the upper side of the finger is connected to the command module using a Two-Wire Interface
(TWI). The IMU module is used for tremor and bradykinesia assessments.
An IMU is attached to the wrist while two force sensor boxes are on both sides of the wrist.
These sensors are used in the rigidity assessment. A medical textile glove incorporates the
command module and the sensor board (IMU). A series of textile gloves are used to fix the
command module and the sensor board to the user’s hand.
The positions of all sensors are shown in Figure 5-3.
System Concept
37
Figure 5-3: Positions of the sensors. For the position of the sensor board (IMU), only one finger from the
index finger and middle finger was chosen according to tremor or bradykinesia tasks.
Figure 5-4: System diagram for an IMU chip. Two Wire Serial Interface (TWI) is compatible with the
Philips’s IIC protocol.
The system diagram of an IMU chip is shown in Figure 5-4. It includes a gyroscope and an
accelerometer.
Figure 5-5: Two parts of the glove monitoring system.
At first, two separate parts were implemented as Figure 5-5 shows, because rigidity assess-
ment is different from tremor and bradykinesia assessment. One part was the system for the
tremor and bradykinesia assessment, while the other was for the rigidity assessment.
5.2.1 System for Tremors and Bradykinesia Assessment
Parkinsonian tremor is called “pill-rolling” tremor, as it is likened to rolling a pill between the
thumb and index finger, because of this, the IMU was placed on the upper side of the index
finger (Giuffrida et al., 2009). For bradykinesia assessment, the inertial sensor in the middle
finger has the best effect. All the data were transmitted to the computer in real-time via a seri-
X-axis
Gyroscope
Y-axis
Z-axis
ADC
ADC
ADCR
eg
iste
rs
X-axis
Accelerometer
Y-axis
Z-axis
ADC
ADC
ADC
Inte
rface
Re
gis
ters
2-wire busMCU
Commnd moduleand sensor board
Computer
Data
USB
Rigidity cuff
Data
Glove
Rigidity assessment part
Tremor/bradykinesia assessment part
System Concept
38
al-to-USB communication interface. The system diagram of the tremor and bradykinesia
quantification system is shown in Figure 5-6.
Figure 5-6: System diagram of the tremor and bradykinesia quantification system. 1) sensor board; 2)
command module; 3) textile glove; 4) universal serial bus (USB) cable; 5) GUI.
5.2.2 System for Rigidity Assessment
The level of parkinsonian rigidity was able to be measured using an IMU and force sensors on
the wrist. Figure 5-7 shows the system diagram of the rigidity assessment system. The rigidity
cuff was connected to the computer via a USB cable.
Figure 5-7: System diagram of the rigidity assessment system (Based on Dai et al., 2013b).
Because the movement of the wrist and elbow has two directions: passive (PA) and contrala-
teral active (CA), both sides of the wrist need a force sensor box.
Each force sensor box included four force sensitive resistors (FSR sensors or FSRs), which
were in parallel connection to one output. The output connected one end to the power supply
and the other to a pull-down resistor to the ground. The point between the fixed pull-down
resistor and the force sensor box was connected to the analog input of a microcontroller.
Compared to a single force sensor, the force sensor box had the benefits of higher measure-
ment stability and a bigger contact patch for the examiner. Two force sensor boxes were con-
nected to the command module. The IMU part (a triple-axis gyroscope and a triple-axis accel-
erometer) was also included in the command module. All the data were transmitted to the
computer via a serial-to-USB communication interface.
Figure 5-8 shows the structure of a force sensor box. Four FSR sensors were located in the
four bottom corners of the housing and were in contact with the rubber feet on its upper side.
This structure has the advantage that the force sensor box provides almost the same value
when an examiner presses on every point of the housing’s upper side with the same force. The
Tremor score:
Bradykinesia score:
ComputerGlove
5
4
2
3
1
1
System Concept
39
viscosity and elasticity of the elbow, which are the major components of mechanical imped-
ance, were calculated with the sensor data and displayed in the GUI (Post et al., 2009).
Figure 5-8: Structure of the force sensor box (Based on Dai et al., 2013a). A force sensor box consisted of
four FSRs in parallel configuration. These four FSRs were located on each corner of the box
Elastic stiffness depends on the torque and angular displacement of the elbow movement (Pat-
rick et al., 2001). If a joint shows viscous behavior, it means that the measured torque de-
pends on movement velocity. In order to avoid modeling the viscous component, some re-
search groups chose to either maintain a constant velocity by using motor actuated systems, or
to advise examiners to impose the same movement on all subjects. This study does not utilize
a motor part in order to keep the device portable nor does it change the clinical assessment
course.
5.2.3 Combined Version
Because the rigidity assessment system was separated from the tremor and bradykinesia as-
sessment system, a new concept with all three symptom assessments is presented. Figure 5-9
shows the system diagram of the combined glove monitoring system which can assess all
primary symptoms. The circuit board and force sensors are embedded into a cuff which is
produced by a 3D printer. There is only a microcontroller in the glove part. The textile glove
is no longer necessary.
Figure 5-9: System diagram of the combined system. All the sensors are connected to a single microcon-
troller. The command module and force sensor boxes are embedded into a case.
Figure 5-10 shows the GUI diagram of the combined system. There are three major parts:
recording lists in the upside of the GUI, start buttons for each task, and progress bars for each
task.
The assessment tasks and algorithms of the separated systems and combined system are still
the same.
System Concept
40
Figure 5-10: GUI diagram of the combined system. The start buttons, progress bars, and recording lists
for each assessment task are listed in the GUI at the same time. Thus it is better for the surgeon to com-
pare the parameters.
5.3 Dynamic System Description
Figure 5-11: Flowchart of the signal processing in the glove monitoring system (Taken from Dai et al.,
2013a). The glove part was based on the sensors and command modules. The computer part was based on
the program in the computer. The communication between them was the serial-to-USB port.
The flowchart of the signal processing methods in the glove monitoring system is shown in
Figure 5-11. The system comprised of two major components: the glove part and the comput-
er part. The sensor data were collected via a microcontroller. The microcontroller sent data to
the computer at a defined time interval.
Timer interruptInitiation
InitiationReceive
sensor data
Dequeue dataData pre-processing
Tremor/Bradykinesia/Rigidity
quantification
Display and
save results
Acquire and send sensor data
Enqueue
data
Glove Part
Computer Part
System Concept
41
The sensor reading was taken by the microcontroller at timed intervals. Gyroscope and accel-
erometer data were simultaneously sampled in order to get high-quality position coordinate
information. Low-pass filtering (LPF) was employed in all sensor outputs for the reason of
anti-aliasing measures.
Inside the glove module, the sensor data were obtained and sent to the computer via a USB
interface. Inside the computer, the received data were stored in a queue. The last cycle’s data
was de-queued for analysis at the same time. Three signal processing methods realized the
three assessment tasks correspondingly. After each assessment task, the parameters according
to the UPDRS score were listed in the recording list. The maximum and minimum values of
the parameters were kept at the bottom of the recording list automatically.
Table 5-1 shows the quantitative models for the objective quantification of neuromotor symp-
toms. These models are based on estimation theory or other statistical analyses.
Table 5-1: Quantification algorithms of the neuromotor symptoms (PD and ET)
TASK PARAMETERS AND METHODS
Tremors Amplitude (Linear regression model)
Bradykinesia Mean and standard deviation of grasp ranges (Statistic analysis)
Rigidity Elasticity and viscosity (Least-squares parameter estimation)
For the rigidity task, viscous and elastic components should be calculated before the calcula-
tion of mechanical impedance.
The computer received the data for continuous signal processing, storage, and display. Data
input and analysis took place in a multi-thread mode, with the use of a queue.
5.3.1 Processing Methods of Tremor Amplitude and Frequency
The algorithms used to quantify the severity of tremors are very important. Some researchers
have proposed objective methods to detect and quantify the tremor severity.
According to the state of the art, Fourier-based spectral analysis, statistical measure (quadratic
mean), and other computational methods such as neural network or fuzzy classifier are per-
formed with the inertial sensor data for the tremor amplitude quantification (Burkhard et al.,
2002; Narcisa et al., 2011; Patel et al., 2009). However, spectral analysis is used for the ma-
jority of these studies.
In this study, signal processing involves IIR and FIR filters as well as other special algorithms
such as PSD analysis. To detect tremors, the signals are then processed with auto power spec-
trum with a certain time length (3–10 seconds).
PSD Estimation
As Figure 5-12 shows, for PSD estimation, peak power means the power estimation around
the dominant frequency in the power spectrum of sensor signals.
System Concept
42
Figure 5-12: PSD estimation of ten-second inertial sensor signals. fdominant represents the dominant fre-
quency of the signals.
Equation 5-1 shows the calculation of the peak power of inertial sensor signals:
, (5-1)
where * denotes the complex conjugate and N is the number of sample points in the sensor
signals.
The formula of the one-dimensional FFT in Equation 5-1 is described as
, (n=0, 1, 2, …, N-1), (5-2)
where X is the input sequence, N is the number of elements of X, and Y is the transform result.
The frequency resolution is fsample rate/N, while fsample rate is the sampling frequency.
Then the power spectrum is converted into a single-side power spectrum. The power spectrum
magnitude (peak power) has units of the input signal unit-RMS squared. The power estima-
tion units of IMU signals are (°/s)2
/Hz and g2/Hz, respectively. Here g is equal to m/s
2.
Tremor Amplitude Estimation
For range (displacement) analysis, the angular velocity obtained from the gyroscope needs to
be integrated over time only once, but the integration of the acceleration signals requires dou-
ble integration. However, the sensor bias and drifts are integrated as well. Thus, the raw data
from the inertial sensors are used for the tremor quantification.
According to the research conducted by Giuffrida et al. (2009), for the rest tremor and postur-
al tremor in PD, the logarithm of the peak powers’ summation of both power spectrums of
accelerometer and gyroscope data had the highest correlation with UPDRS scores (coefficient
of determination r2=0.9). For the action tremor in PD, the RMS sum of both gyroscope and
accelerometer data had the highest correlation with UPDRS (r2=0.69) (Heldman et al., 2011).
In this system, the logarithm of the peak powers’ summation of the IMU sensor signals is
regarded as the tremor amplitude.
Given its oscillatory nature, tremors are well suited to spectral analysis, the most popular
method for tremor quantification (Cameron et al., 1997). The goal of PSD is to describe the
0 f1
f2
fdominant
Frequency (Hz)
Po
we
r e
stim
atio
nPeak power
System Concept
43
power distribution of a signal based on a finite set of data over the frequency domain.
Quadratic mean (RMS) interprets actual levels, while PSD results show frequencies that con-
tribute the most to the tremor. Because the tremors are based on a dominant frequency, the
advantages of PSD compared to a statistical measure (quadratic mean) are that it highlights
the tremor signals from noise and other movement with analysis in the frequency dimension
and the squared value of the signals. There are nonparametric methods, parametric methods,
and subspace methods for PSD estimation. The nonparametric methods include periodogram,
Welch's method, and the multitaper method (MTM). Parametric methods are used to estimate
the output signal from a linear system driven by white noise. Subspace methods, which are
also known as high-resolution methods, are used to generate frequency component estima-
tions for the signal based on an eigenanalysis or eigendecomposition of the correlation matrix.
The tremor signals will be tested with different types of methods for PSD estimation at a later
time. In addition, for the offline tremor data analysis, PSD results should be used together
with quadratic mean results to identify the overall features of tremors.
The flowchart of the signal processing for the tremor quantification in this study is shown in
Figure 5-13. The sensor data are from the glove part, which is based on the sensor board and
command module. In the computer part, the sensor data are band-pass filtered from 3 to 12 Hz
(tremor band) before tremor amplitude quantification.
Figure 5-13: Signal processing for the tremor detection and quantification in a single finger (four chan-
nels).
Accelerometer X, Y,Z Signal
Drift Cancellation
RMS Calculation and
Remove Gravity Component
Auto Power Spectrum(RMS)
Power & Frequency Estimation
Power Peak & Dominate
Frequency (Pa;Fa)
Gyroscope X,Y,Z Signal
Drift Cancellation(X,Y,Z)
Band-pass Filter(X,Y,Z)
Auto Power Spectrum(X,Y,Z)
Power & Frequency EstimatIon
(X,Y,Z)
Power Peak & Dominate
Frequency
(Pgx,Pgy,Pgz;Fgx,Fgy,Fgz)
P=Max(Pa,Pgx,Pgy,Pgz)
Dominate Frequency F=0 P>0.05
F=Fa
F=Fgx
F=Fgy
F=Fgz
Estimated Power: P' around FDominate Pole: P', F
*Scaling Factor
*Scaling Factor
Band-pass Filter
P=Pa
N Y P=Pgx
P=Pgy
P=Pgz
Y
Y
Y
Y
System Concept
44
Initially, the signals from the gyroscope and the accelerometer in a single finger are analyzed
separately. The single-sided, scaled, and auto power spectra of the signal of each gyroscope
axis and the combined values of all three axis outputs of the accelerometer, of which length is
set to three seconds, are separately computed to get the four-channel power spectra in real-
time. The dominant frequencies and the estimated peak powers of these frequencies are thus
obtained. Then the estimated peak powers will be multiplied by scaling factors. The channel
with the highest peak power is the dominant channel, and its dominant frequency is the domi-
nant frequency (F) of all channels. Other channels are needed to perform PSD estimations
again with the dominant frequency, and multiply the peak powers by the scaling factors again.
The power estimations in all axes around the same dominant frequency (F) are calculated and
sum up to P'.
The power spectrum of every axis in a certain period (three seconds) is calculated continuous-
ly. The dominant frequency and peak power can be obtained as soon as the tremor occurs.
When the dominant frequency is between 3.5 Hz and 7.5 Hz and the sum of peak powers is
more than a threshold, the tremor is reported.
However, the signals in ten seconds are used for the PSD estimation at the end of the ten-
second tremor assessment task. Then, the sum of peak powers, which is regarded as the trem-
or amplitude, and the dominant frequency can be displayed in the recoding list.
PSD estimation is the most popular method for tremor amplitude calculation, because the
tremor is a rhythmic and involuntary movement based on a dominant frequency (Patel et al.,
2009). Gyroscope and accelerometer react respectively to rotational and linear movements.
Then a linear regression model is used to fit the clinical ratings (UPDRS tremor scores and
TETRAS) and the peak powers from both the gyroscope and accelerometer signals.
The total output of a triple-axis accelerometer can be expressed as axyz:
222
zyxxyz aaaa . (5-3)
axyz also includes gravitational acceleration, which equals 9.81 m/s2
in vector product. Gravita-
tional acceleration can be removed from the accelerometer outputs with high-pass filters.
Then there is only one axis acceleration data for the following signal processing.
The dominant frequency can be calculated using PSD estimation. If the dominant frequencies
in different axes are not the same, the valid dominant frequency in the axis with the highest
peak power is defined as the dominant frequency of all axes. Then the total peak power in all
four axes, which includes axyz and three-axis gyroscope signals, is the power of all axes’ data
around the valid dominant frequency with the PSD method. The peak power in all axes after
normalization is regarded as the amplitude of tremor. Heldman et al. presented the discovery
that the logarithm of the peak power in all triple-axis accelerations and three-axis angular ve-
locities correlates well with the clinical scores of tremor (Burkhard et al., 2002).
Because the accelerometer and gyroscope are used to measure linear and rotational movement
respectively, the accelerometer presents a higher correlation for some tremor tasks, while the
gyroscope performs a higher correlation for other tremor tasks. Then a multiple linear regres-
sion model is used to fit the clinician ratings (UPDRS tremor scores) and the peak powers
during each tremor task (Timmer et al., 1997):
Then the linear regression model (Giuffrida, et al., 2009) can be expressed as
System Concept
45
R = R0+ ln (b0·PAxyz+cx·PGx+cy·PGy+cz·PGz), (5-4)
where R is the predicated tremor score; R0, b0, cx, cy, and cz are the regression coefficients;
PAxyz, PGx, PGy, and PGz are the peak powers for the triple-axis accelerometer and triple-axis
gyroscope, respectively. R0 and the three scaling factors are different for the three tremors
(rest, posture, and action).
For different tasks and different tremor types, these regression coefficients are different. The
peak power in this study is the power estimation around the dominant frequency with ± 0.3
Hz length in the single sided power spectrum of ten-second sensor signals (Heldman et al.,
2011).
These three regression coefficients were obtained according to Stevens’ power law in psycho-
physics (Luce & Krumhansl, 1988). Table 5-2 shows the initial coefficients and scaling fac-
tors for different tremor tasks.
Table 5-2: Coefficients and scaling factors of the tremor amplitude regression models
Tremor Coefficients
R0 b0 cx cy cz
Rest 0.8 10 0.001 0.001 0.001
Posture 0.6 5 0.0001 0.0001 0.0001
Action 0.3 2 2e–5 2e–5 2e–5
In the future, the parameters in Table 5-2 will be modified according to the results of meas-
urements in patients with tremor.
The occurrence of tremor in a patient depends on many factors. Tremor can disappear some-
times even for a patient with severe tremor. Therefore, it is important to quantify tremor se-
verity during the stable tremor state. The tremor state is classified into two types in this pro-
ject: valid state and invalid state. The signals in invalid state will be discarded.
After a ten-second tremor assessment task, the tremor signals need to be checked. Valid state
means the stable state in both the time domain and frequency domain. The judgments of valid
state are listed as follows:
In the frequency domain of ten-second signals, the proportion of peak power to the
whole power estimation should be bigger than 85%.
In the time domain, the standard deviation of ten-second angular velocity ranges
(peak-to-peak values of all axes of the gyroscope) should be smaller than 30% of the
mean gyroscope signal ranges.
5.3.2 Processing Methods of Bradykinesia Parameters
Hand grasping is easier to perform during DBS surgery. The signals of healthy people have a
consistent amplitude and frequency, thus appearing sinusoidal. On the other hand, patients
with severe bradykinesia have an inconsistent amplitude and frequency.
System Concept
46
Signal processing for bradykinesia quantification is done by the computer. After receiving the
sensor data from the microcontroller, the gyroscope signals are filtered firstly in real-time.
The flowchart of bradykinesia detection and quantification using a gyroscope is shown in
Figure 5-14.
As shown in Figure 5-14, the angular displacement during hand grasping can be calculated by
numerical integration of the triple-axis angular velocities at the end of the middle finger. Then
the mean value and standard deviation value ( and ) of hand grasping ranges ( ) can
be acquired by statistical methods. After a ten-second assessment, a second-order integration
is performed with all the gyroscope signals. Then a peak-detection algorithm and statistical
analysis are performed.
Figure 5-14: Flowchart of the signal processing for bradykinesia quantification in the computer (Based on
Dai et al., 2013a). Here peak-to-peak angles, which are calculated with a peak detection algorithm, mean
the peak to peak values of all hand grasping cycles.
Figure 5-15 shows the peak-detection method during bradykinesia quantification.
Figure 5-15: Peak detection for grasping ranges in a bradykinesia task. There are five hand grasping
cycles in this figure. A peak-detection algorithm could be used to calculate peak to peak angle values
(Ranges 1 to 5: 1 to 5). An angular displacement threshold, for example ±20º for both maximum
and minimum, can be used to remove the unnecessary peak points.
The grasping ranges ( PPα or ) are the three-dimensional peak-to-peak values during grasp-
ing cycles of a bradykinesia task. At first, the triple-axis grasping ranges are calculated sepa-
System Concept
47
rately. The combined triple-axis grasping range ( ) is the sum of the three-axis grasping
ranges ( ).
The number of peak to peak angle values during the ten-second assessment period is related
to the dominant frequency of hand grasps. The mean value of peak to peak angle values
(grasp ranges) represents the amplitude of bradykinesia. The standard deviation value of grasp
ranges represents the change of amplitude during the grasping task.
The mean and standard deviation of hand grasping ranges are easy to calculate with statistical
methods. The dominant frequency of hand grasps is calculated from the gyroscope signals
using the FFT method.
The observed parameters are:
Dominant frequency of hand grasping movement (f).
Mean range in hand grasps ( ).
Standard deviation value of hand grasp ranges ( ).
These parameters are calculated based on the gyroscope signals from the middle finger. For a
patient with mild bradykinesia, the signals obtained from the gyroscope should have a con-
sistent amplitude and frequency and should appear sinusoidal. Conversely, the signals from a
patient with severe bradykinesia should have a much lower and inconsistent amplitude and
frequency. Thus the mean value and SD value of the hand grasping ranges and hand grasping
frequency during a bradykinesia assessment task represent the bradykinesia severity (Zhang et
al., 2011). The grasp cycles during a bradykinesia assessment task (ten seconds) are approxi-
mately equal to ten times the dominant frequency of the hand grasping movement (f).
Equation 5-5 and Equation 5-6 show the mathematical formulas used to determine and
:
, (5-5)
, (5-6)
where is the combined hand grasping range (peak-to-peak values) in a single grasp cycle,
and N is the number of the hand grasping cycles in a ten-second bradykinesia task.
Grasping Angle Calculation
There are three methods for the hand grasping angular displacement (grasping range) calcula-
tion:
Triple-axis tilt angle calculation based on the gravity accelerations from a triple-axis
accelerometer.
Triple-axis angular velocity integration calculation with the triple-axis gyroscope sig-
nals.
System Concept
48
Six-axis sensor fusion with the signals from a triple-axis gyroscope and triple-axis ac-
celerometer.
Tilt calculation is a static measurement. The gravity is used as an input to determine the orien-
tation of an object calculating the tilt degree.
For the angle calculation with the triple-axis gyroscope signals, first order approximation
(trapezoidal method), Simpson’s rule or bodes rule can be utilized.
For sensor fusion, the gyroscope is used as the primary source of orientation information. The
accelerometer is used for roll-pitch drift correction.
It is not easy to separate the gravity components from the accelerometer data as the hand
grasping is not static movement. A Kalman filter or the DCM algorithm is used to combine
the data from the gyroscopes and accelerometers to produce an output that is better than that
obtained from individual sensors. However, sensor fusion (DCM or Kalman filter algorithm)
makes the program more complicated, because several parameters need to be set and there are
several seconds of adaptive time before the stable state is reached in the program (Madgwick
et al., 2011).
Second-order integration approximation (Simpson’s method) is easy to perform, and the drift
in ten seconds has a small effect on the angle calculation. Then the grasping range is calculat-
ed via triple-axis angular velocity integration and combination (National Instrument Inc.,
2011). The single-axis angular displacement (α i) in i-th sampled point is calculated according
to Equation 5-7 (Simpson’s method).
i
1j
1jj1ji vv4v6dta , (5-7)
where vj is the angular velocity in j-th sampled point and dt is the sample interval (0.02 s for
50 Hz sampling rate).
Because there are three axes in the signals of hand grasps, the hand grasp angles should be
calculated in three axes separately.
5.3.3 Processing Methods of Rigidity Parameters
At first, the IMU and force sensor boxes should be calibrated and verified.
Mechanical impedance is a mathematic model used to describe the dynamic features of the
linear vitiation system. For a stable linear-vibratory system, its outputs (displacement, veloci-
ty, and acceleration), under a simple harmonic and alternating force are the harmonic move-
ments with the same frequency. There is, however, a delay in phase. Mechanical impedance is
a feature of frequency response. The system outputs depend on the features of the system such
as inertia, velocity, stability, and stiffness.
The viscous and elastic components, which are the main components of the mechanical im-
pedance, are modeled in this study (Charles, 2003).
Rigidity assessment is performed by the measurement of elbow angle movement and torque
on the wrist (Patrick et al., 2001). The outputs of two force sensor boxes and an IMU around
the wrist area are:
System Concept
49
αa ,,, 21 FF , (5-8)
where and are the outputs of force sensor box 1 and force sensor box 2, respectively; a
and are the acceleration and angular velocity of elbow movement, respectively. Their units
are N, g, and rad/s, respectively.
The triple-axis elbow angles (α) during the rigidity task can be calculated from the IMU out-
puts ( αa , ) in real-time using the DCM algorithm (Madgwick et al., 2011).
The calculation of elastic stiffness (c) and viscosity (d) is realized by using a least squares
parameter estimation method (regression analysis) to solve Equation 5-9 with ten-second data
(Patrick et al., 2001).
The torque (T) in a rigidity assessment task can be expressed as:
edcl)FF(T 21 αα , (5-9)
where l is the length of the subject’s arm, c is the elastic stiffness of the subject’s elbow, d is
the viscous stiffness (viscosity) of the subject’s elbow, and e is the constant offset of the sen-
sors. The arm length (l) must be determined before the measurement, and set in the system’s
user interface through keyboard entry.
With ten-second duration and 100 Hz sampling rate, a 1000-point data array with the format
of Equation 5-8 is obtained during a single rigidity assessment task. Equation 5-9, which is
the model of the elbow joint movement, can be written as:
α α . (5-10)
Therefore, Equation 5-10 can be written in vector form as:
, (5-11)
where the components of Equation 5-11 are:
α α (5-12)
(5-13)
After the angular displacement calculation of the elbow’s passive movement, the angular ve-
locities and elbow angular displacements are combined in matrix A. Angular velocity ( ) is
the first column in matrix A. On the condition that the elbow moves exactly in one direction,
the three-dimensional angular displacement and angular velocity of elbow movements can be
replaced with single-axis data points in the Equation 5-8. In addition, the combination of all
sensors and axes needs further signal proceeding, such as fuzzy analysis, classification and
regression trees, artificial neural networks, support vector machines, or other pattern recogni-
tion techniques.
Thus, a least squares estimation is obtained using the following equation:
. (5-14)
Mechanical impedance is the feature of the parkinsonian rigidity, and is calculated as follows:
System Concept
50
, (5-15)
where f is the frequency of hand movement and is obtained with a peak-detection algorithm
from the angular displacement of elbow movement; and fd 2 is the modified viscous
stiffness.
In order to acquire the relation between mechanical impedance and the UPDRS ratings, the
relation of elastic stiffness and viscosity with the rigidity severity should first be investigated.
Six-Axis Direction-Cosine-Method Algorithm
Direction cosine method is another way to construct a rotation matrix, other than Euler angles
(Edwan et al., 2011).
The six-axis DCM algorithm is based on a triple-axis gyroscope and a triple-axis accelerome-
ter. The gyroscope is used as the primary source of orientation information. The accelerome-
ter is used for roll-pitch drift correction because it has no drift over time. Only the gravity
vector of the accelerometer is used for the drift detection.
Figure 5-16: Block diagram of DCM algorithm. Here PI controller denotes a proportional-integral con-
troller. For six-axis DCM algorithm, there is no yaw-axis input for the drift detection.
As shown in Figure 5-16, a proportional plus integral (PI) controller is used to control the
drift adjustment. Each of the rotational drift correction vectors (roll and pitch) is multiplied by
weights and fed to a PI feedback controller. Then the drift adjustments are added to the gyro-
scope vectors to produce corrected gyroscope vectors. The outputs of the algorithm are three-
dimensional angles (orientation).
α α
α
a
Compass or GPS
Drift adjustment Integration & Normalization Orientation
Drift detection
Gravity detection
PI controllerError
Gravity
dt.
Gyroscope α
Adjustment
Patch & Roll
Yaw
Accelerometer
α
Prototypical Realization
51
6. Prototypical Realization
At first, the realization of the nine-axis DCM algorithm is presented in Chapter 6.1. In addi-
tion, two tremor and bradykinesia assessment systems with wireless communication inter-
faces, which have already been implemented at the beginning of this study, are presented in
Chapter 6.2. These systems could be used in patients’ homes and in future versions.
After that, prototypical realizations of the tremor/bradykinesia assessment system and rigidity
assessment system are presented. The materials, hardware configuration, and software de-
scription of these two systems are presented in Chapters 6.3, 6.4, and 6.5, respectively. As
described in Chapter 6.6, inertial sensors and force sensor boxes of these systems were cali-
brated according to the sensor calibration procedures.
At last, the realization of the combined system, which can assess the three primary symptoms,
is presented in Chapter 6.7.
6.1 Nine-Axis Direction-Cosine-Method Realization
As Figure 5-16 in Chapter 5.3.3 shows, signals of GPS (global positioning system) or a triple-
axis compass can be used to reduce yaw drift. Such a multi-sensor fusion method, which in-
cludes the signal processing of a triple-axis compass, a triple-axis gyroscope, and a triple-axis
accelerometer, is a nine-axis DCM algorithm.
The nine-axis DCM algorithm can be realized in an eight-bit MCU of small size. A prototype
based on the nine-axis DCM algorithm and an eight-bit MCU has been implemented.
Figure 6-1: Atmel Inertial Two Sensor Board (AVRSBIN2) from Atmel Inc., USA (© Atmel, 2011).
As shown in Figure 6-1, an Atmel Inertial Two Sensor Board (AVRSBIN2), which delivered
a full nine-DOF inertial sensor platform, was used to supply the sensor fusion algorithm. This
circuit board combined an accelerometer (KXTF9, Kionix Inc., USA), a compass
(HMC5883L, Honeywell Inc., USA), and a gyroscope (IMU-3000, InvenSense Inc., USA).
Another circuit board, which included a microcontroller (ATMEGA644, Atmel Inc., USA)
and other electronic components, was used to run DCM algorithm.
Prototypical Realization
52
The sampling rates of both the gyroscope and accelerometer were 50 Hz, while the sampling
rate of the compass was 10 Hz. The gyroscope measured rapid movement, while the accel-
erometer and compass removed the drift in the gyroscope signal.
The three-axis angles, which were the results of sensor fusion using the DCM method, were
sent to a computer in 50 Hz. As shown in Figure 6-2, a 3D demo displayed the rotational
movement of the sensor board (AVRSBIN2). This demo program was modified based on the
program of Julio et al. (2009).
Figure 6-2: Nine-DOF AHRS demo based on Python 2.6.4. Python is a simple but powerful open-source
programming language.
6.2 Prototypes with Wireless Communication Interfaces
Two prototypes with wireless communication for tremor/bradykinesia assessment are pre-
sented in this section. A prototype was for the motion tracking of the wrist and all fingers,
while the other for the motion tracking of a single finger.
At first a glove-based system with several accelerometers and gyroscopes was used to meas-
ure the hand’s motion. As shown in Figure 6-3, the desired hardware included sensors, a re-
chargeable battery, and a microcontroller. These electronic components were embedded in a
small case in the same fashion as a wrist blood pressure monitor. The case incorporated con-
nectors to the sensor module. The sensors were able to be attached and removed from the fin-
gers without difficulty.
The IMU3000 gyroscope by Invensense Inc., with the dimensions of 4mm 4mm 0.9mm,
and the SMB380 accelerometer (3mm 3mm 0.9mm) by BOSCH, and the MMA8452Q
accelerometer (3mm 3mm 1mm) by Freescale were chosen for use in this design.
Prototypical Realization
53
These inertial sensors were placed on the upside of a washable conductive textile (Textronics
Inc.). Five of these conductive textiles were able to be attached on each finger of a hand. The
conductive textile on the thumb was connected to the power supply (level “1”), while the
conductive textiles on other fingers were connected to the circuit ground respectively via a
resistor (level “0”). The communication between the microcontroller (ATMeaga644) and sen-
sors was achieved with a two-wire serial interface (TWI) multiplexer (PCA9546A, TI, USA).
Two NanoLOC AVR Modules (microcontroller and 2.4 GHz radio transceiver, Nanotron
Technologies GmbH, Germany) were used for the communication between the microcontrol-
ler and the computer. All the electronic components in the wrist were embedded into the cuff
of a wrist blood pressure monitor and powered by a lithium-ion battery (3.7 V, 1400 mAh).
The NanoLOC AVR Module included an Atmel AVR microcontroller, which managed the
sensors and sent the data to a computer via wireless communication. Another NanoLOC AVR
Module, which was plugged into a computer, received the data and sent it to the computer via
RS232 to USB transmission.
a) b) c) d)
Figure 6-3: Components of the measurement system with wireless communication. a) NanoLocAVR Mod-
ule (35mm × 14mm × 3mm); b) lithium-ion battery; c) touch sensor based on a washable conductive tex-
tile; d) the cuff from a blood pressure monitor (Beurer BC19, Beurer GmbH, Germany). A NanoLocAVR
module, lithium-ion battery, and other additional components were placed inside the cuff of this blood
pressure monitor.
Figure 6-4 shows the hand motion measurement system and the sensor module which can
measure the movements in all fingers and the wrist.
a) b)
Figure 6-4: Hand motion measurement system with sensor modules (inertial sensors) and touch sensors.
a) hand motion tracking system; b) schematic of a touch sensor. Each touch sensor’s output was connect-
ed to a digital input pin of the microcontroller. The sensor module can be attached to the fingers and con-
nected to the command module
Touch sensor
(ground)
Sensor
modules
Touch sensor
(outputs)Touch sensor (ground)
Touch sensor
(output)
Conductive
textile
Prototypical Realization
54
Figure 6-5 shows a prototype with the wireless communication interface. Rest, postural, and
action tremor tasks, together with the finger taps task, can be assessed by this system. One
disadvantage of this system is that it cannot be used in an operating room.
Figure 6-5: Wireless tremor and bradykinesia assessment monitor with a single sensor module. Its case
was also based on a blood pressure monitor (Beurer BC19, Beurer GmbH).
6.2.3 Graphical User Interfaces
Figure 6-6 describes the GUI of the hand motion measurement system for tremor and
bradykinesia assessments with four sensor modules, which were located on the wrist and three
fingers. This system could be used for bradykinesia and tremor assessments.
Figure 6-6: GUI of the hand motion tracking system (for finger tapping task). Finger 1 in this figure
means the index finger. This GUI corresponds to hardware in Figure 6-4.
The raw data from each axis of the gyroscope and the combined value of all three axes’ out-
puts of the accelerometer, together with the touch sensor signals, were displayed in waveform
mode.
The dominant frequencies and the estimated powers of two fingers’ movements were comput-
ed and displayed directly. The auto power spectrum of each finger movement was displayed
at all times. When the dominant frequency and its estimated power met the definition (fre-
quency range and amplitude threshold), the tremor indicator would light up. The display was
refreshed at 0.05 second intervals.
To save the dominant frequency and power values, the key labeled “Acquire Data” was
pressed.
Prototypical Realization
55
Finger tapping and supination-pronation movements were used to assess bradykinesia based
on the hand motion measurement system. The change in angles and the time taken for the two
fingers to come into contact and separate were displayed in real-time. The raw data of all sen-
sors were stored in the computer in real-time.
Figure 6-7 represents the GUI of playback mode for this measurement system. The stored raw
data were able to be opened and replayed with this user interface.
Figure 6-7: Review of the stored sensor signals in a tremor or bradykinesia assessment task of the hand
motion tracking system (playback mode). This GUI corresponds to the hardware (Figure 6-4) and soft-
ware (Figure 6-6).
Figure 6-8 and Figure 6-9 show the GUIs of the prototype in Figure 6-5, which only had a
sensor board and a wireless communication interface. These GUIs were based on LabVIEW
2010 Evaluation Version (National Instruments Corp, USA) and Visual C# 2010 (Microsoft
Inc., USA) respectively.
Figure 6-8: LabVIEW-based GUI of the wireless tremor and bradykinesia assessment monitor with a
sensor board. This GUI corresponds to the wireless tremor and bradykinesia assessment system in Figure
6-5.
Prototypical Realization
56
Figure 6-9: Visual C#-based GUI of the wireless tremor and bradykinesia assessment monitor with a sin-
gle inertial sensor module. This GUI also corresponds to the wireless tremor and bradykinesia assessment
system in Figure 6-5 (© TUM-MIMED, 2013).
As can be seen from Figure 6-8 and Figure 6-9, the built-in graphical user interface compo-
nents, such as buttons and graphs, make LabVIEW more intuitive for displaying signals and
results. It is also easy to modify the LabVIEW program. Thus, the realization of a test version
using LabVIEW has some advantages. The program, which is based on Visual C#, is more
readable and is better to be embedded in a commercial system considering its lower price.
6.3 Materials
6.3.1 Selection of Sensors
The first step in implementing the glove monitoring systems was to select the appropriate sen-
sors.
Sensor Specifications
According to the literature (Giuffrida, et al., 2009; Heldman, et al., 2011a; Patrick, et al.,
2001), the full-ranges of sensors in the neurological symptom assessments are listed as fol-
lows:
Angular velocity: ± 2000 º/s (degrees per second or dps) in three dimensions.
Acceleration: ± 4 g in three dimensions; here g is the gravitational acceleration (1g =
9.8 m/s2).
Attached force on both sides of the wrist: 0–80 N; here N is Newton (1 Newton
=101.97 grams).
MEMS IMUs and FSR sensors were chosen in this project for their small dimensions and
low-cost compared to other types of sensors.
Figure 6-10 shows the inertial sensors involved in this study.
Prototypical Realization
57
Figure 6-10: Inertial sensors used in this study (IMU3000 /MPU6050 /MPU6150 /MPU9150 /MMA8452Q
/SMB380) (© Invensense, 2013; © Freescale, 2012; © Bosch, 2010).
The Standardized Sensor Performance Parameter Definitions for MEMS sensors, which was
organized by Intel, Qualcomm, and the MEMS Industry Group (MIG), was not released until
May 2013. Therefore, not all of the primary parameters of inertial sensors were available in
the datasheets.
The primary features of available inertial sensors and IMUs are listed in Table 6-1. These
parameters were taken from the relative datasheets. These sensors are from manufactures such
as Invensense, Bosch, ST, Kionix, and ADI, respectively.
Table 6-1: Comparative specifications of inertial sensors. Here “⁄” means no embedded sensor fusion;
“–” means no available information; Gyro. refers to a 3-axis gyroscope and Acc. denotes a 3-axis ac-
celerometer.
Sensor
Embedded
sensor
fusion
Package
Current (mA)
[Gyro.; Acc.]
(total)
RMS Noise
[Acc.; Gy-
ro.]
Resolution
(bits)
[Gyro.; Acc.]
Release
date
IMU3000
Digital
Motion
Processor
440.9
mm; QFN 6.1
0.1°/s rms,
0.01º/s/√Hz 16 05/2010
SMB380 ⁄ 330.9
mm; QFN 0.2 500 µg/√Hz 10 09/2007
MPU6050
Digital
Motion
Processor
440.9
mm; QFN 3.6; 0.5; (3.9)
400µg/√Hz;
0.05°/s rms 16;16 11/2010
MPU6150
Digital
Motion
Processor
440.9
mm; QFN 3.6; 0.5; (3.9)
400µg/√Hz;
0.2°/s rms 16;16 11/2010
LSM330 iNEMO 33.5
1mm; LGA 6.1; 0.25 – 16; 16 07/2012
BMI055 ⁄ 34.50.95
mm; LGA (5.15)
150µg/√Hz;
0.014°/s/√Hz 16; 12 09/2012
KXG02 ⁄ 440.9m
m; LGA
3.75; (4.0)
150 µg/√Hz 16; 16 01/2012
ADIS163
67 ⁄
232323
mm; ML-
24-2
49 0.5mg//√Hz;
0.044°/s/√Hz 12; 12 01/2010
An IMU works by detecting both the current rate of acceleration (using a 3-axis accelerome-
ter) and the changes in rotational attributes (using a triple-axis gyroscope). At the beginning
Prototypical Realization
58
of this study, a triple-axis MEMS gyroscope and a triple-axis MEMS accelerometer were used
to constitute an IMU. After the availability of single-chip IMUs, MPU6150 and MPU6050
were chosen in this project for the following reasons:
Their pin-out and package (QFN) are compatible with the IMU3000, which we had
used before the availability of IMU chips. MPU 6000 and MPU6150 were first availa-
ble on the market as single-chip IMUs.
A chip with QFN package is better to solder than the chip with LGA package in labor-
atory conditions.
Their outputs are digital, obtained via a TWI/ SPI (serial peripheral interface) interface
by the MCU.
MPU6150 is actually a lower cost version of the MPU-6050 and is focused on television re-
mote control and gaming applications. MPU-6050 (or MPU-9150) has better performance
with regard to bias offset, drift rate, and noise levels.
Figure 6-11 shows the system structures of MPU6150 and MPU9150, which has an additional
compass layer.
Figure 6-11: Structure of the multi-sensor integration (© Invensense, 2013). Six-axis MPU6150 has a die of
two layers, while the nine-axis MPU9150 has a three-layer die.
Figure 6-12 shows the system diagram of MPU6150, which is same as MPU6050.
Figure 6-12: Block diagram of MPU6150/MPU6150.
Calibration
Sensor RegistersSensor RegistersSensor RegistersSensor
Registers
FIFO
Digital Motion Processor
(DMP)
2-wire SerialInterface
MPU6X50
TWI
LDO3.3VX Accel.
Y Accel.
Z Accel.
X Gyro.
Y Gyro.
Z Gyro.
Temp. sensor
ADC
ADC
ADC
ADC
ADC
ADC
ADC
ADC
Sig
na
l Co
nd
ition
ing
Prototypical Realization
59
A gyroscope has much higher power consumption than an accelerometer or a compass in
working state, more specifically, 3.6 mA for a triple-axis gyroscope compared to 0.2 mA for a
three-axis accelerometer inside the Invensense MPU6050 or MPU6050.
Figure 6-13 shows the axis orientations of MPU6050 and MPU6150.
Figure 6-13: Axis orientations of MPU6150 and MPU6050. The benefit of this structure is that the axes of
the accelerometer and gyroscope are the same. Thus the adjustment of axes before sensor fusion is not
needed.
The main part of a force sensing resistor (FSR) is a conductive polymer, which detects physi-
cal pressure or force applied in the upside. FSRs are simple to use due to a low cost and small
thickness. The disadvantages are their low precision (from 5% to 25%) and possible damage
by continuously being attached with pressure (for several hours). The major manufacturers of
FSRs are Interlink Electronics Inc., Tekscan (Flexifore sensors), and IEE International Elec-
tronics & Engineering Inc., etc. The information regarding the FSRs, which are suitable for
this project, is listed as Table 6-2.
Table 6-2: Comparative information of FSRs. Here “–” means no available information. For the
FSR402, the best force range is from 0 to 20 N, but it can be extended to 100 N when applying force
evenly over the active area (0.125 cm2) of its surface area.
Sensor Range
(N)
Repeat-
ability
Linearity
(Error)
Hystere-
sis
Response
time Active area
Current
density
FSR402 0–20N* +/- 2% – +10% 3 ms Ø12.7 mm < 1 mA
FSR149N
S 0–100N ± 3% – 20% 2–3 ms Ø6 mm < 1 mA/cm
2
FSR152 0–100N ± 3% – 20% 2–10 ms Ø15.2 mm < 1 mA/cm2
A401 0–110N ±2.5% ±3% 4.5 % 5µs
(5 ms) Ø25.4 mm < 2.5 mA
These sensors are robust polymer thick film (PTF) sensors that exhibit a decrease in resistance
to an increase in force applied to the surface of the sensors. They are with analog outputs. The
output of an FSR should be connected to the ADC pin of a microcontroller.
One disadvantage of the FSRs is that they can provide only single-axis analog output, but the
elbow movement during the rigidity assessment task is a three-dimensional motion. MEMS
three-axis force sensors have appeared with digital interface but as of yet there is not a fully-
developed product available for our project. For example, a capacitive three-axis force sensor
(WEF-3A) from Wacoh Inc., Japan, is very small (Ø: 10mm, H: 7mm) and has a built in 32-
bit ADC, but the operating current is 200 mA. In the future, MEMS three-axis force sensors
can be used in this project.
Prototypical Realization
60
Figure 6-14 shows the structure of a round FSR. Figure 6-15 shows the photo of FSR149NS
and FSR 402 short.
Figure 6-14: Structure of a round FSR (© Interlink, 2012).
Figure 6-15: FSR149NS and FSR 402 short.
6.3.2 Additional Components
As Figure 6-16 shows, a series of TG®
medical gloves (L&R GmbH, Germany) can be worn
by different patients. These gloves have the advantage of being washable, hard-wearing com-
fortable, and have high durability. They are also CE marked and sterilizable. In addition, they
meet the sterilization standards in DIN EN ISO 11135, 11137, and 17665.
Figure 6-16: Medical gloves made by Lohmann & Rauscher TG, with four different sizes.
The microcontroller selected for the tremor/bradykinesia assessment system was an
ATMega1284P, while the microcontroller selected for the rigidity assessment system was an
ATMega328. Both these chips were manufactured by Atmel Inc., USA.
The missing sampling data have a very bad effect on the PSD results (Spieker et al., 1995).
The bit rate (or call baud rate) for the serial communication was set to 115200 bps (bits per
second) or higher. The crystal oscillator frequency is better at 7.3728 MHz, 11.0592 MHz,
18.432 MHz or 22.118 MHz. Thus, the bit error rate of serial communication between the
MCU and command module is about zero.
Prototypical Realization
61
The speed grades of both the ATMega1284P and ATMega328 are:
0 10 MHz @ 2.7 5.5V.
0 20 MHz @ 4.5 5.5V.
The crystal oscillator frequencies of the two prototypes for the tremor/bradykinesia assess-
ment and the rigidity assessment were higher than 10 MHz. Therefore, their power supplies
were chosen to be +5 VDC.
FT232R (Future Technology Devices Inc., UK) is a USB to serial UART (universal asyn-
chronous receiver/transmitter) interface which was to be used here.
Figure 6-17: Alarm control cable LiYY 4 × 0.14 mm².
As Figure 6-17 shows, a shielded four-pin cable was used to connect the sensor board to the
command module.
6.4 Physical Implementation
Figure 6-18: Internal architecture (hardware) of the glove monitoring system.
MPU6150(Gyro_ /Accel_)
MCU(ATMEGA1284)
TWI
Voltage converter (LDO 3.3V)
AVR_USB JTAGConnector
Sensor board
Command Module
JTAG
USB
5V
data
Power
Forcesensors
Rigidity cuff
MCU(ATMEGA328)
Voltage converter (LDO 3.3V)
ISP Connector
MPU6150(Gyro_ /Accel_)
MPU6150(Gyro_ /Accel_)
MPU6150(Gyro_ /Accel_)
Prototypical Realization
62
Figure 6-18 shows the hardware diagram of the glove monitoring system, which included two
parts. The command module was for the tremor and bradykinesia assessment, while the rigidi-
ty cuff was for the rigidity assessment.
6.4.1 Prototype for Testing
At first, the command module was supposed to be integrated into the textile glove. As shown
in Figure 6-19, the command module was fixed in the glove with a durable elastic fabric Vel-
cro strap. A USB 2.0 type A to Mini-B five-pin cable was used to connect the command
module to a computer.
Figure 6-19: Photo of the textile-integrated command module and mini USB connector.
One disadvantage of this design was that the circuit part of the command module could come
into contact with the human body. Another disadvantage of the design was that the USB con-
nection was not stable.
Thus the command module was placed in a case for reasons of safety. According to the exper-
iments by Lorenzo et al., a USB JTAG cable (Lorenzo et al., 2011) was chosen for its better
connection stability.
6.4.2 Tremor/Bradykinesia Assessment System
Figure 6-20: Internal architecture of the tremor/bradykinesia assessment system.
Prototypical Realization
63
As Figure 6-20 shows, the command module and the sensor board were used for finger mo-
tion data acquisition and communication with a computer. The command module acquired
sensor data and sent it to the computer. An IMU (MPU6150) was located on the sensor board.
The implementation of the sensor board is shown in Figure 6-21 and Figure 6-22. The sensor
board (IMU) required a +3.3 V power supply and a TWI bus. Then a four-pin cable connected
the sensor board with the command module.
Figure 6-21: Schematic of the sensor board.
Figure 6-22: Photo of the sensor board (dimensions: 14mm × 11mm × 2mm).
As Figure 6-22 shows, the sensor board was put into a liquid insulation can (PLASTI DIP
LIQUID TAPE, Plasti Dip® GmbH, Germany) for insulation protection. The board was in-
side the can for one minute and then taken out to cool for three hours.
Figure 6-23 shows the circuit board and the housing of the command module. Three screws
were placed in the top side of the housing. The implementation of the command module is
shown in Figure 6-24 (a).
a) b)
Figure 6-23: Photos of the circuit board and housing of the command module. a) command board; b)
housing with four screws (M2 × 6mm). There was also a prototype with only three screws.
Prototypical Realization
64
Figure 6-24 shows the command module with the sensor board and a USB JTAG cable. The
USB JTAG cable, which includes a serial-to-USB chip (FT232R, Future Inc., UK), was con-
nected to a female 30-pin connector.
a) b)
Figure 6-24: Photos of the command module and USB JTAG cable. a) photo of the command board with
the sensor board; b) USB JTAG cable for the communication between the command module and a com-
puter.
There were five connectors on the circuit board of the command module. A female 30-pin
connector was used to connect the command module to a computer via a JTAG USB cable.
The JTAG interface of the microcontroller was connected to this connector for programming.
A four-pin connector was used for the connection with the sensor board. Two connectors
were available for the force sensor boxes but were not used in this version. The USB port of
the computer provided the power supply (+5 V) for the command module. The voltage was
regulated from 5.0 V to 3.3 V with a low dropout linear regulator (XC6204B332MRN, Torex
Inc., Japan).
The IMU in the sensor board was connected to the microprocessor (ATMega 1284P, Atmel
Inc., USA) via the TWI interface. The microprocessor sampled sensor data at 100 Hz or 50
Hz. A green light-emitting diode (LED) also indicated the state of sensor readings at 100 Hz
or 50 Hz.
The final implementation of the tremor/bradykinesia assessment system is shown in Figure 6-
25.
a)
b)
Figure 6-25: Overview of the prototype for tremor/bradykinesia assessment. The command module was
34mm × 32mm × 7mm. The command module and sensor board both had indirect contact with the human
body.
USB JTAG cable
Command module
Textile glove
Sensor board
Prototypical Realization
65
The operation of the wired system for tremor/bradykinesia is shown in Figure 6-26.
Figure 6-26: Tremor/bradykinesia assessment system.
6.4.3 Rigidity Assessment System
The rigidity assessment system was implemented before the availability of single-chip IMUs
such as MPU6150/6050. Therefore, two inertial sensors were adopted. A triple-accelerometer
(MMA8452Q, Freescale Inc., USA) worked with a triple-gyroscope (IMU3000, Invensense
Inc., USA).
Two differential force sensor boxes were used to measure the attached force on the wrist.
Each force sensor box was built from four FSRs (FSR-149NS, IEE Inc., Luxembourg), which
were connected in parallel. Each force sensor box had only one output, with a type of voltage
divider.
Figure 6-27 shows the realization of a force sensor box. The output of the force sensor box
(Vout) was directly connected to the ADC input of a microcontroller (ATMega 328p, Atmel
Inc., USA).
a) b)
Figure 6-27: Implementation of a force sensor box. a) schematic of the force sensors; b) photo of the force
sensor box (Based on Dai et al., 2013a).
A plastic case was used to place the force sensor boxes and can be pressed by an examiner
during rigidity assessment. The rigidity cuff consisted of a plastic housing, a Velcro textile
band, and a microprocessor board.
Figure 6-28 shows the Velcro textile band. Through a combination of hard Velcro and soft
Velcro, the two force sensor boxes and the microcontroller board were fixed on the wrist. The
microcontroller board and the housing of the two force sensor boxes were all attached on the
Prototypical Realization
66
outside of the band. The textile band had the advantage of being able to fit the different arm
diameters of each patient.
Figure 6-28: Photo of the Velcro textile band (© TUM-MIMED, 2012).
Because the power supply of the microcontroller (5 V) differed from that of the inertial sen-
sors, a dual bi-directional TWI bus voltage-level translator (PCA9306, TI Inc., USA) was
used. Figure 6-29 shows the schematic diagram of the voltage-level translation circuit.
Figure 6-29: Schematic of the dual bi-directional TWI bus voltage-level translator.
Figure 6-30 shows the circuit board of the rigidity cuff. The implementation of the command
module is shown in Figure 6-31. All the sensor data were sampled at 100 Hz by the micropro-
cessor.
Figure 6-30: Photo of the microcontroller board in the rigidity cuff.
Figure 6-31: Overview of the prototype for rigidity assessment (Based on Dai et al., 2013b).
The prototypes of both the tremor/bradykinesia and rigidity assessment system are shown in
Figure 6-32.
Prototypical Realization
67
Figure 6-32: Prototypes of the glove monitoring system. 1) command module (tremor/bradykinesia as-
sessment part); 2) rigidity cuff (rigidity assessment part). The command module and sensors did not have
direct contact with the human body (Taken from Dai et al., 2013a).
6.5 Graphical User Interface Implementation
The data transmission, signal processing, and GUI of the tremor/bradykinesia assessment sys-
tem were carried out on a computer using LabVIEW 2010 Evaluation Version (National In-
struments Corp., USA). Figure 6-33 shows the GUI of the bradykinesia and tremor assess-
ment system.
Figure 6-33: GUI of the tremor/bradykinesia assessment system. 1) serial interface setting; 2) raw data
path configuration; 3) start button; 4) progress bar; 5) results recording list; 6) real time result area; 7)
sensor raw data; 8) results recording chart; 9) PSD chart.
The GUI included the task start buttons, progress bars, recording lists, and the storage loca-
tion setting of raw data.
To start the tremor assessment, the task button in the GUI must be set to “Tremor” and the
examiner needs to press the start button. During tremor assessment, the raw data and the PSD
figure of three-second signals were displayed at the bottom. At the end of the tremor task, the
actual assessed results, which included dominant frequency and tremor amplitude during the
ten-second assessment task, were displayed in the result area of tremor assessments.
To start bradykinesia assessment, the task button in the GUI must be set to “Kinesia”. During
bradykinesia assessment, the raw data and the PSD figure as well as the angular displacement
were displayed at the bottom. At the end of the bradykinesia task, the actual assessed results,
1
2
1
2
3
4
5
6
7
8
9
Prototypical Realization
68
which included dominant frequency, mean and standard deviation of the grasping ranges,
were displayed in the result area of bradykinesia.
After each tremor or bradykinesia assessment task, the corresponding assessment result item,
which consists of the assessment number and relative parameters, was added to the results
recoding list automatically. The result items with the biggest and smallest amplitudes were
displayed at the bottom of the results recoding list. All the raw sensor data were stored in the
computer for offline analysis.
Data communication and the GUI of the rigidity assessment system were realized with Visual
C# 2010 (Microsoft Inc., USA). This program invokes MATLAB 2012b (MathWorks Inc.,
USA) to perform signal processing. Figure 6-34 shows the GUI of the rigidity assessment
system.
Because the patient’s forearm length could not be measured directly with the above sensors,
the examiner should record the subject’s forearm length in the input box of the GUI.
Figure 6-34: GUI of the rigidity assessment system (Baed on Dai et al., 2013b). 1) raw data path configura-
tion; 2) serial interface setting; 3) start button; 4) input box for the patient’s forearm length; 5) progress
bar; 6) communication status; 7) results.
6.6 Calibration of Inertial Sensors and Force Sensor Boxes
The calibration methods for inertial sensors are first introduced in Chapter 6.6.1. The sensors
in the rigidity cuff were first calibrated. However, the calibration of the tremor/bradykinesia
assessment system depends on each IMU on the sensor boards. There are several sensor
boards available. Each IMU should be calibrated respectively.
6.6.1 Analysis of Inertial Sensors
The key issue in the performance of a motion tracking system is its accuracy. Although pre-
sent inertial sensors are smart sensors (factory calibration and run-time calibration firmware
are embedded into the sensors); bias, noise, scaling factor error, and the cross-axis misalign-
ment compensation matrix still require consideration. Some initial calibration procedures for
individual gyroscopes and accelerometers are needed.
Prototypical Realization
69
Gyroscope Calibration
The single axis output (angular velocity ) of a gyroscope in time t is expressed as fol-
lows:
, (6-1)
where is the true angular velocity, a and b are the scaling and biasing of the output, re-
spectively. The output bias error (offset) is related to its special structure, temperature change,
magnetic field, and other noises. Linear acceleration can also result in bias in a gyroscope. In
general, this bias error results in big drifts after an integration time.
Thus the first-order triple-axis gyroscope compensation equation is shown as follows:
, (6-2)
where Sgx, Sgy, and Sgz are the scaling factors in three axes; , , and are the offsets
in three axes; and M is the 3×3 cross-axis misalignment compensation matrix.
In this study, the offsets are calculated with the mean gyroscope outputs in a static state in ten
seconds. The calculation function is described as follows:
, (6-3)
where , , and are the raw data outputs in static state; N is the number of sampled
points (= sampling rate × 10 seconds).
The true angular velocities ( , , and ) from a gyroscope are considered to be linear
according to the datasheets of gyroscopes. The sensitivity of a single gyroscope axis is com-
puted with the values recorded while rotating the gyroscope 180° around that axis. The angu-
lar displacement by integrating the angular velocity is a sinusoidal signal. Those values are
integrated over time and multiplied by 1/(sample rate). The signals of each gyroscope axis are
then divided by the scaling factor to obtain the angular velocity in [°/s]. Then the scaling fac-
tors in three axes are described as:
, (6-4)
where is the gyroscope’s sensitivity scaling factor [unit: count/(°/s)], which can be
found in the datasheet; and is the sampling interval (unit: s).
Accelerometer Calibration
Accelerometer offsets, scaling factors, and cross-axis misalignment errors should also be
compensated by the calibration process.
Thus the first order triple-axis gyroscope compensation equation is shown as follows:
Prototypical Realization
70
, (6-5)
where Sax, Say, and Saz are the scaling factors in three axes; , , and are the offsets
in three axes; and M is the 3×3 cross-axis misalignment compensation matrix.
Figure 6-35: Positions of an IMU for calibrating accelerometer offsets.
The outputs of a triple-axis accelerometer include linear accelerations and gravitational accel-
erations. For the calibration of static offsets, the sensor is in a stable condition before being
measured. Each axis of the accelerometer is placed parallel to gravity with both positive and
negative sides. As Figure 6-35 shows, the true value of the related acceleration in the static
state should be 1g or -1g. However, the outputs of the accelerometer include also bias (off-
sets). The recorded gravitational acceleration both with and against gravity, as shown in Fig-
ure 6-35, are described as follows:
(6-6)
The triple-axis accelerometer outputs are scaled to multiples of the gravitational constant (g).
These values are used to calculate the sensitivity and offset of each axis. Then the offsets can
be calculated with the values in Equation 6-2 as follows:
=
(6-7)
The scaling factors can be calculated with the values in Equation 6-2 as follows:
=
(6-8)
where is the sensitivity (unit: count/g) of the accelerometer.
+g -g +g -g
+g -g
Prototypical Realization
71
The sensor outputs should be sampled at room temperature and the sensors should have been
working for at least one hour before the measurement.
6.6.2 Calibration of Inertial Sensors
According to the calibration procedures in Chapter 3.6.3, the calibration results of the gyro-
scope (IMU3000) and the accelerometer (MMA8452Q) in the rigidity cuff are shown from
Table 6-3 to Table 6-5.
Table 6-3 Offsets and sensitivities of the gyroscope in the rigidity cuff (IMU3000, 16-bit resolution: -
3276832767)
X-Axis Y-Axis Z-Axis
Offset -390 +161 +68
Sensitivity 0.9358 0.9491 0.9611
Table 6-4 Accelerometer outputs when its axes were oriented along/against the direction of gravity
(MMA8452Q, 12- bit resolution: -20482047)
X-Axis Y-Axis Z-Axis
Orientation along g-axis (1g) 991 1011 1008
Orientation against g-axis (-1g) -1005 -974 -997
Table 6-5 Offset and sensitivity for the accelerometer (MMA8452Q, 12-bit resolution: -20482047)
X-Axis Y-Axis Z-Axis
Offset -7 +18 +5
Sensitivity 0.9746 0.9697 0.9785
6.6.3 Calibration of Force Sensor Boxes
According to Figure 5-8 in Chapter 5.2.2, the output of the force-to-voltage conversion was
tied to a measuring resistor in a voltage divider configuration. The output voltage of a force
sensor box (Vout) was shown as:
MFSRccout RRVV /1/ , (6-9)
where Vcc was the power supply (+5 V); RFSR was the resistor value of the force sensor box;
and RM was the value of the pull-down resistor (4.7 KΩ).
Equation 6-9 shows that the output of a force sensor box response for RFSR is similar to an e-
function. The force-resistance characteristic of a single FSR sensor (FSR149NS) is nonlinear
and the response approximately follows an inverse power-law characteristic. However, a line-
ar force-to-voltage relationship is needed in Equation 5-9 in Chapter 5.3.3. Hence, a regres-
sion analysis was applied to get the voltage-force function. Regression analysis was per-
formed with the Curve Fitting Toolbox in MATLAB R2012a (Mathworks Inc., USA). Results
of a two-term Gaussian regression were compared to one-term and four-term Gaussian func-
tions.
Prototypical Realization
72
Since the FSR output is nonlinear and sensor sensitivity varies, every force sensor box has to
be respectively calibrated. As each force sensor box included four FSR sensors but had only
one output, the calibration of FSRs was performed after the implementation of the two force
sensor boxes.
As shown in Figure 6-36, 58 weight loads ranging from 5 grams to 6000 grams were applied
at the center of the top side of each force sensor box. Since FSRs are prone to drift, only the
value up to three seconds after applying the load could be used and the weight was released
from the force sensor box after every single measurement. Therefore, 116 voltage values were
acquired. The correlation of sensor outputs over weights was recorded.
Figure 6-36: Setup of the force sensor box’s calibration. Each force sensor box was tested with weights,
while the weights were placed in the center top side of the force sensor box (Based on Dai et al, 2013a).
If the applied force was lower than 0.5 N, there was no change in the output of the force sen-
sor box. However, for applying forces bigger than 0.5 N, the resistance of the force sensor
box rapidly dropped and reached saturation when the applied force was more than 100 N. The
FSR sensor tutorial (IEE International Inc., Luxemburg, 2011) states that the FSR sensor re-
sponse to higher loads is similar to an e-function. Therefore, a Gaussian function was chosen
for regression analysis:
, (6-10)
where x was the output of the force sensor box and f(x) was the attached force.
To evaluate the performance of the Gaussian function, results were compared with fits for an
e-function and polynomial functions (Abramowitz et al., 1965).
Figure 6-37: Two-term Gaussian regression plot of force sensor box 1: weight (g) versus the output of the
force sensor box 1 (© TUM-MIMED, 2012).
1) Weight
2) Force sensor box
1
2
Force sensor box output (V)
Weig
ht (g
ram
)
0 0.5 1.0 1.5 2.0 2.5
fit
weight vs force sensor box ouput
Prototypical Realization
73
Figure 6-38: Two-term Gaussian regression plot of force sensor box 2: weight (g) versus output of the
force sensor box 2 (© TUM-MIMED, 2012).
Using the MATLAB Curve Fitting Toolbox, one-, two-, and four-term Gaussian, e-Function,
and Polynomial regression analyses were performed with the measurement data.
Figure 6-37 and Figure 6-38 show the regression plots of a two-term Gaussian regression
function which fitted to the two force sensor box outputs, while Figure 6-39 shows a one-term
Gaussian regression fitting and Figure 6-40 shows a four-term Gaussian regression fitting of
the force sensor box 1.
Figure 6-39: One-term Gaussian regression plot of force sensor box 1: weight (g) versus output of the
force sensor box 1 (© TUM-MIMED, 2012).
Figure 6-40: Four-term Gaussian regression plot of force sensor box 1: weight (g) versus output of the
force sensor box 1. The red circle highlights the area with over-fitting (© TUM-MIMED, 2012).
Force sensor box output (V)
We
ight
(gra
m)
0 0.5 1.0 1.5 2.0 2.5 3.0
fit
weight vs force sensor box ouput
Force sensor box output (V)
We
igh
t (g
ram
)
0 0.5 1.0 1.5 2.0 2.5
fit
weight vs force sensor box ouput
Force sensor box output (V)
Weig
ht (g
ram
)
0 0.5 1.0 1.5 2.0 2.5 3.0
fit
weight vs force sensor box ouput
Prototypical Realization
74
As Figure 6-39 shows, one-term regression did not fit the response curve well, producing low
values of correlation (r2) (see Table 6-6). At the same time, the four-term regression was sus-
ceptible to outliers, which resulted in over-fitting (Figure 6-40).
Table 6-6 shows the results of the different curve fits applied to the data set. The selected fit
has been highlighted. The MATLAB Curve Fitting toolbox was not able to find a one-term
Gaussian fit for the dataset of force sensor box 2. Accordingly, the corresponding value in
Table 6-6 is “no fit”. In addition, there were no four-term e-Function fittings for the two force
sensor boxes.
Table 6-6: Results of regression analysis: adjusted r2 values (force sensor box 1/ force sensor box 2)
Gaussian e-Function Polynomial
1-term 0.9904 / no fit 0.9896 / 0.9925 0.8908 / 0.7778
2-term 0.9937 / 0.9971 0.9931 / 0.9923 0.9826 / 0.9531
4-term 0.9934 / 0.9973 Not available in toolbox 0.9939 / 0.9961
As shown in Table 6-6, a two-term Gaussian function has the highest correlation with the
measured data:
,
(6-11)
where F(vout) was the calculated force value, vout was the sensor box output voltage, and a, b,
and c were coefficients.
The coefficients of Equation 6-11 could also be determined using the MATLAB Curve Fitting
Toolbox. These parameters, which have been calculated using the two-term Gaussian regres-
sion, are given in Table 6-7.
Table 6-7: Coefficients of the two-term Gaussian regression for the two force sensor boxes (FSR box-
es)
a1 b1 c1 a2 b2 b3
FSR box 1 5.91e+04 1104 387.1 1020 322.6 195.5
FSR box 2 1.139e+18 7287 1159 734.6 445.6 253.4
6.7 Combined Version
Together with the tremor/bradykinesia assessment system and the rigidity assessment system,
a portable monitoring system used to quantify all primary neurological symptoms in PD and
ET was able to be implemented.
As shown in Figure 6-41, the combined system was built up based on the two prototypes
(tremor/bradykinesia assessment system and rigidity assessment system).
Some components were replaced for the reason of a higher performance. The crystal oscillator
frequency was set to 18.432 MHz instead of 11.0592 MHz. MPU6050 replaced MPU6150 for
its better gyroscope noise performance. The active area of FSR149NS is too small (Ø4.03
Prototypical Realization
75
mm) and therefore each force sensor box needed four force sensors. FSR402 has a 14.7 mm
diameter active area coverage, thus only one force sensor was needed in the combined ver-
sion.
Figure 6-41: Internal architecture of the combined version.
Figure 6-42 shows the connection between the command module and the force sensor boxes.
The force sensor boxes were connected to the microcontroller via an instrumentation amplifi-
er.
Figure 6-42: Interfaces between the force sensor boxes and command module (MCU board). The outputs
of the two force sensor boxes were connected to an amplifier. TLV2372 (TI Inc., Dallas, USA) is a rail-to-
rail input/output low power instrumentation amplifier.
For the tremor/bradykinesia assessment system and rigidity assessment system, the received
sensor data in the computer were checked with an imitated tremor test by healthy subjects.
The test result indicates that the missing points of sampling data (dropped packets) over the
whole sampled data were less than 0.2% with a sampling rate of 50 Hz. However, sometimes
there were disconnections between the command module and the computer. The assessment
task was able to be interrupted by violent hand movement. In the combined version, one side
of the USB cable was soldered directly to the printed circuit board of the command module,
instead of using a USB connector in the command module.
Prototypical Realization
76
A 10-pin header (FTSH-105-01-L-DV-K-A-P, Samtec Inc., Indiana, USA) was placed on the
center of the command board as the JTAG programming connector, because of its small di-
mension and safe style (keyed).
Figure 6-43 shows the hardware of the combined version.
Figure 6-43: Prototype realization of the combined version.
Figure 6-44 shows the GUI which can be used to assess tremor, bradykinesia, and rigidity.
Figure 6-44: GUI of the combined version (Taken from Dai et al., 2013a).
6.8 Conclusion
The prototype of the tremor/bradykinesia assessment system is presented. It was based on an
IMU, which was attached on a finger, and could be combined with a textile glove.
The first prototype of a portable rigidity assessment system, which was attached to the wrist,
was also presented. It was based on two force sensor boxes and an IMU. As a result, this sys-
tem made it easy to perform passive elbow movement.
The combined version was implemented for quantification of all primary neurological symp-
toms in patients with PD or ET.
For possible hardware design in the future, a microcontroller with an embedded USB com-
munication interface can provide a higher transmission rate between the computer and micro-
Prototypical Realization
77
controller than the serial-to-USB converter. In addition, a system with a wireless interface
instead of wired communication makes the system usable outside of the operating room and
in places such as homes or sickrooms. As a result, the system can be used for bradykinesia
assessment at home or in hospital for further tests.
The signal processing was mainly performed by the computer. The role of the microcontroller
was simply to acquire and send the raw data to the computer.
The Kalman filter and DCM algorithm can be run on a microcontroller. As mentioned in the
first section of this chapter, the nine-axis DCM algorithms can be run on an eight-bit micro-
controller. However, the microcontroller needs more time consummation in this situation.
Thus the sampling rate of the sensors is limited to a low value. In order to perform certain
timing issues such as signal pre-proceeding by the side of the command module, two micro-
controllers may be used: one as the master for sensor data acquisition and the other as the
slave for signal processing. In addition, the sensor fusion can be embedded inside the sensor,
which performs an orientation calculation. Then the angular displacement of tremor, grasping
angles during a bradykinesia task, and elbow angles during a rigidity task can be easily ac-
quired. The angles from the chip-based sensor fusion should be compared to the values ac-
quired with the methods in this system.
Passive elbow movement is a movement in three dimensions. However, the force sensor box-
es provided the different output only in a single axis. Thus the passive movement should be
parallel to one axis of the Cartesian coordinate system.
Tremor/bradykinesia assessment systems with wireless communication interfaces can be used
as mobile medical devices used before and after the DBS surgery. These prototype realiza-
tions are helpful for designs in the future.
Experiments and Discussions
78
7. Experiments and Discussion
IMU do not give position information, but most research projects use raw IMU data for signal
processing in tremors and bradykinesia quantitative assessments. No previous studies have
compared the assessment systems for neurological symptoms based on IMU with other medi-
cal motion tracking systems.
Before comparing this system’s output with the patients’ clinical ratings, it is necessary to
evaluate the calculated technical parameters of this system. In order to do this, several exper-
iments were carried out. Analytical verifications of the three test tasks of the glove monitoring
system were performed. The verifications of analytical methods are presented in Chapters 7.1,
7.2, and 7.3, respectively.
After verifying the system, the tremor and bradykinesia assessment system was tested with
tremor and bradykinesia tasks in the hospital. The experiments of tremor and bradykinesia
assessments are presented in Chapter 7.5 and Chapter 7.6, respectively.
7.1 Verification of Analytical Methods for Tremor Assessment
The NDI Aurora electromagnetic tracking system is first introduced in Chapter 7.1.1. It was
used for the verification of analytical methods for tremor, bradykinesia, and rigidity assess-
ments.
The verification of tremor amplitude and frequency is presented in Chapter 7.1.2.
7.1.1 NDI Aurora Electromagnetic Tracking System
As shown in Figure 7-1 (a), an Aurora electromagnetic (EM) spatial measurement system
(Northern Digital Inc., Waterloo, Canada) was used as the reference system. The Aurora® EM
system is a navigation system designed specifically for medical applications.
As shown in Figure 7-1 (b), the Aurora’s micro six-DOF EM sensor has a small dimension
(2.5 mm L11 mm, line: 2 m) and can be attached to a finger or wrist.
a) b)
Figure 7-1: a) Aurora electromagnetic (EM) tracking system; b) Aurora six-DOF sensor cable tool (©
NDI, 2013).
The Aurora system’s characterized measurement volume can be set to a dome volume (480
mm L660 mm) or cube volume (500mm 500mm 500mm). However, there is a 50 mm
distance between the planar field generator and the characterized measurement volume. Fig-
6 DOF sensor
Planar field generator
System control unit
6 DOF sensor
Sensor interface unit
Experiments and Discussions
79
ure 7-2 shows the characterized measurement volumes. The EM tracking in cube volume
mode has higher accuracy than the dome volume mode. Thus the measurement volume was
fixed to cube volume mode in this study. The accuracy parameters of the EM tracking system
with a six DOF sensor are:
Location: 0.48 mm (RMS), 1.40 mm (95% confidence interval).
Orientation: 0.30º (RMS), 0.48º (95% confidence interval).
In this study, the sampling rate of the EM tracking system was 40 Hz. The IMUs of the glove
monitoring system were sampled at 100 Hz.
Figure 7-2: Characterized measurement volume (left: cube volume mode; right: dome volume mode).
Figure 7-3 shows the experimental setup of all verifications of analytical methods.
Figure 7-3: Setup for the verification of analytical methods. An electromagnetic tracking system was run
with the glove monitoring system at the same time. The graphical user interfaces of both systems were run
on two laptops. The data transmissions of the two systems were both based on serial-USB communication.
7.1.2 Verification of Tremor Amplitude and Frequency
Motivation
IMU signals are angular velocity and linear acceleration; but a surgeon judges tremor severity
according to the displacement of hand tremor. On the other hand, the EM tracking system can
obtain position information directly. Therefore, the comparison of the analytical methods be-
tween EM signals and IMU signals was performed. Here the glove monitoring system refers
to the tremor/bradykinesia assessment system.
Laptop for
EM tracking system
Laptop for
glove monitoring system
Glove monitoring system
and EM sensors
Glove monitoring system
and EM sensors
EM generator
EM control unit
EM sensor unit
1)
2)
3)
4)
5)
6)
1
2
3
5 4
6
Experiments and Discussions
80
Hypothesis
By comparing with those from the EM system, the tremor amplitude (peak powers) and dom-
inant frequency of the glove monitoring system should meet the following requirements:
Mean value and standard deviation of the differences between dominant frequencies:
fmd < 1.00±0.88 Hz (Niazmand et al., 2011b).
Correlation coefficient of peak powers (tremor amplitude): r >0.95 (Bland & Altman,
2003).
Materials
NDI Aurora® EM tracking system with a six-DOF sensor (Northern Digital Inc., Can-
ada)
Tremor/bradykinesia assessment system (included a JTAG USB cable and a command
module with a sensor board)
Laptop installed with the application software of the EM system (Aurora Toolbox)
Laptop with the application software of the tremor/bradykinesia assessment system
(LabVIEW-based GUI for tremor/bradykinesia assessment; MATLAB R2008b for
data analysis)
Method
The dominant tremor frequencies calculated by the tremor/bradykinesia assessment system
and EM system were ftb and fem (see Chapter 5.3.1). Then the difference of their dominant
frequencies in a single tremor task was represented by:
(7-1)
Therefore, fmd (mean and standard deviation of dominant frequency differences) can be calcu-
lated by the values of fd (frequency differences) during all imitated tremor assessment tasks:
± (
, (7-2)
where N was the amount of imitated tremor assessment tasks.
The tremor amplitude (peak powers based on the IMU signals) calculated by the
tremor/bradykinesia assessment system was R, while the tremor amplitude (peak power based
on the position data) judged by the EM system was E (see Chapter 5.3.1). Then the correla-
tion coefficient between these two parameters was:
, (7-3)
where N is the amount of imitated tremor assessment tasks.
Experimental Setup
As shown in Figure 7-4, together with the IMU in the tremor/bradykinesia assessment system,
a six-DOF EM sensor was attached to the subject’s middle finger for real-time finger motion
tracking. Because of the limited EM tracking dimension, only the postural tremor task was
Experiments and Discussions
81
performed by nine healthy subjects (average age: 29±2.1 years). Each subject performed sim-
ulative postural tremors four times with different amplitudes (no tremor, slight, moderate, and
severe). The glove monitoring system and Aurora EM system were running at the same time
during the experiments. The raw sensor data from the EM sensor and IMU were respectively
stored in the two laptops in real-time.
Figure 7-4: Setup for the verification of tremor quantification (Taken from Dai et al., 2013a). The test
subject performed “pill-rolling” action of the hand.
Figure 7-5: Raw signals measured with both the EM and tremor/bradykinesia assessment systems during
an imitated tremor task (Taken from Dai et al., 2013a). a) triple-axis EM localization signals (unit: mm);
b) IMU signals: triple-axis gyroscope signals and a one-axis combined acceleration signal (gyroscope unit:
°/s; accelerometer unit: g).
Figure 7-5 shows the ten-second signal waveform of both the EM sensor (localization signals)
and IMU (angular velocity and acceleration). According to Equation 5-4 in Chapter 5.3.1, the
acceleration readings in three axes were combined into an absolute linear acceleration and
then the gravitational acceleration was partly removed.
Results and Discussion
The parameters obtained from the tremor/bradykinesia assessment system were compared to
those calculated from the positional data, which were the outputs of the EM system, using the
IMU
EM sensor
EM generator
a)
b)
Experiments and Discussions
82
PSD estimation method. The peak powers, which is regarded as tremor amplitude, and domi-
nant frequencies from the two systems were compared.
The dominant frequencies between the two systems had little difference in the range from 2 to
6.5 Hz, even when the subjects did not perform a movement with stable tremor frequency.
As shown in Figure 7-6, for the postural tremor task, the dominant frequency of EM data is
plotted against the dominant frequency of the IMU signals from the tremor/bradykinesia as-
sessment system. The correlation coefficient (r) of dominant frequency between these two
systems was 0.996.
Figure 7-6: Correlation of the dominant frequencies in the imitated tremor tasks between the EM and
tremor/bradykinesia assessment system (unit of the dominant frequency: Hz). The frequencies were calcu-
lated using the PSD estimation method (Based on Dai et al., 2013a).
The maximum, mean, and standard deviation of the frequency differences between these two
systems were 0.570 Hz, 0.115 Hz, and 0.144 Hz respectively. Therefore, fmd (0.115±0.144
Hz) was smaller than 1.00±0.88 Hz.
A linear correlation between the two systems’ peak powers (tremor amplitude) could not be
shown. The reason for this is believed to be that seven subjects did not realize the imitated
tremor activity with a consistent amplitude and frequency during each timed task.
According to the experimental results, inconsistent movement made the peak power of the
EM signals smaller when using the PSD method. When some sampling points in the EM sig-
nals were missed, the peak power increased. These two points are important for the calcula-
tion of tremor amplitude.
For the two subjects who performed a stable movement (i.e., means consistent amplitude and
frequency as shown in Figure 7-5) the correlation between the two peak powers of the glove
monitoring system and the EM tracking system is shown in Figure 7-7. The peak power of the
glove monitoring system was calculated with both gyroscope and accelerometer signals using
the PSD method, while the peak power of EM signals was calculated only from the position
signals of the subject’s finger. Figure 7-7 shows that the correlation between these two sys-
tems was approximately linear. The correlation coefficient between these two systems
(r=0.97, p<0.001) was larger than 0.95.
0 1 2 3 4 5 6 70
1
2
3
4
5
6
7
Dominant frequnecy calculated from EM data
Do
min
an
t fr
eq
un
ecy c
alc
ula
ted f
rom
IM
U d
ata
Difference between frequencies:
ftb
(H
z)
fd
fem – ftb relation
0 1 2 3 4 5 6 70
1
2
3
4
5
6
7
Dominant frequnecy calculated from EM data
Do
min
an
t fr
eq
un
ecy c
alc
ula
ted
fro
m I
MU
da
ta
Difference between frequencies:
: fem (Hz)
Experiments and Discussions
83
Figure 7-7: Relationship of the tremor amplitudes (peak power) acquired with both the EM and trem-
or/bradykinesia assessment system (Based on Dai et al., 2013a). Two subjects each performed four imitat-
ed tremor tasks. The imitated tremor amplitudes varied in range from slight to severe for each subject.
For future research of the glove monitoring system, the RMS values of all sensor data should
be calculated. Together, the RMS values and peak powers could be used to determine the
tremor amplitude. The relation between the peak power of IMU signals and the consistency of
IMU signals in PD patients should be studied in the future.
The basic changes in the EMG or MER signals caused by PD symptoms are increased tonic
background activity and an alternating pattern of EMG or MER bursts. During the EMG or
MER analysis of PD symptoms, special attention has been paid to the analysis of these bursts
by measuring their counts, magnitudes, durations and frequencies (Rissanen et al., 2007). For
vigorous movement measurement, IMU provides good results. However, for slight motion
disorders, the difference between EMG and IMU requires further verification.
7.2 Verification of Hand Grasping Angle Calculation
Motivation
As the bradykinesia quantification is based on hand grasping angles, the hand grasping angle
calculation based on the integration of the triple-axis gyroscope signal is verified in this sec-
tion. Here the glove monitoring system refers to the tremor/bradykinesia assessment system.
Hypothesis
By comparing with those from the EM tracking system, the angular displacements and domi-
nant frequency of the hand grasps measured by the glove monitoring system should meet the
following requirements:
Mean value and standard deviation of the differences of dominant frequencies: fmd <
1.00±0.88 Hz (Niazmand et al., 2011b).
Mean difference of the angular ranges of hand grasps: md < 0.1 · em.
Here em is the mean angular range of hand grasps (peak-to-peak value) measured by the
EM tracking system.
Materials
An NDI Aurora®
EM tracking system with a six-DOF sensor (Northern Digital Inc.,
Canada)
-10 0 10 20 30 40 50 60-15
-10
-5
0
5
10
Peak Power from EM data: (mm2/Hz)
Norm
aliz
ed P
eak P
ow
er
fr
om
IM
U d
ata
:
E
R
Experiments and Discussions
84
A tremor/bradykinesia assessment system (included a JTAG USB cable and a com-
mand module with a sensor board)
A laptop with the application software of the EM system (Aurora Toolbox)
A laptop with the application software of the tremor/bradykinesia assessment system
(LabVIEW-based GUI for tremor/bradykinesia assessment; MATLAB R2008b for
data analysis)
Parameters
Difference of the dominant frequencies from both the EM tracking system and the
tremor/bradykinesia assessment system
Mean difference of the hand grasping ranges between the EM tracking system and
tremor/bradykinesia assessment system during a ten-second hand grasping task
The standard deviation value of the differences of the hand grasping ranges from both
the EM tracking system and tremor/bradykinesia assessment system during a ten-
second hand grasping task
Methods
The dominant frequencies of hand grasps calculated by the tremor/bradykinesia assessment
system and the EM tracking system were ftb and fem respectively (see Chapter 5.3.2). Then the
difference of their dominant frequencies was:
. (7-4)
As described in Equation 7-2, fmd (mean and standard deviation of frequency differences)
could be calculated with the values of fd during all bradykinesia assessment tasks.
The mean angular range of hand grasps calculated by the tremor/bradykinesia system and the
EM tracking system were tb and em respectively (see Chapter 5.3.2). Then the difference
of their mean angular ranges of hand grasps during a single bradykinesia task was:
. (7-5)
Therefore, md could be calculated with the values of d during all bradykinesia assessment
tasks:
, (7-6)
where N was the amount of imitated bradykinesia assessment tasks (N=18 in this experiment).
Experiment Setup
As Figure 7-8 shows, the tremor/bradykinesia assessment system and the EM tracking system
were used to measure the hand grasps by nine healthy subjects (average age: 29.0±2.1 years).
Each test subject performed such tasks three times. The test subject’s hand was in the meas-
urement volume of the EM tracking system. A six-axis EM sensor and a sensor board (IMU)
were attached to the middle finger.
Experiments and Discussions
85
Figure 7-8: Setup for the verification of the bradykinesia quantification method.
Results and Discussion
As shown in Figure 5-14 in Chapter 5.3.2, only the gyroscope signal was used to calculate the
bradykinesia parameters. Thus, the errors in the calculated grasping angles were mainly com-
posed of the gyroscope drift and integration error. To reduce the errors in a short period of
time, the gyroscope was calibrated upon system initiation. Figure 7-9 shows the raw data of
the EM tracking system and the glove monitoring system for a ten-second bradykinesia task.
Figure 7-9: Raw sensor data both from the EM tracking system (location and orientation data) and the
glove monitoring system (angular velocity) during the one-time bradykinesia task of a subject (Taken
from Dai et al., 2013a).
According to the analysis of the sensor signals of this experiment, three subjects performed
hand grasping movements with different frequencies. Therefore, their data were removed.
The EM signals (three-dimensional orientation data) were processed using the FFT method to
get the dominant frequency and the peak-detection method to obtain grasping ranges. The
IMU signals (angular velocity) were processed using the methods described in Chapter 5.3.2.
Table 7-1 shows the differences of bradykinesia parameters between the EM tracking system
EM sensor & IMU
αxa
ya
za
0 1 2 3 4 5 6 7 8 9 10-500
0
500
IMU
angula
r velo
city
(
degre
e/s
)
αza
yaxa
Time (s)
100
0
-1005
-100
10100
0
0 1 2 3 4 5 6 7 8 9 10
0 1 2 3 4 5 6 7 8 9 10
500
0
-500
EM
orie
nta
tio
n
(de
gre
e)
EM
lo
ca
tio
n
(mm
)
IMU
an
gu
lar
ve
locity
(de
gre
es/s
)
Experiments and Discussions
86
and the IMU-based glove monitoring system (six test subjects, each performed the task three
times).
Table 7-1: Absolute differences of parameters between the EM tracking system and the IMU-based
glove monitoring system (bradykinesia task)
Parameters Frequency (Hz) Mean range: SD of ranges:
Max. Difference 0.70 22.18° 7.43°
Mean Difference 0.18 17.60° 3.65°
SD Difference 0.30 6.24° 3.14°
The difference of dominant frequencies (fmd) between the glove monitoring system and the
EM tracking system (0.18±0.30 Hz) was smaller than 1.00±0.88 Hz.
The angular ranges of the finger movements were more than 200° in a single cycle during this
experiment. Thus the difference in mean range between these two systems ( md =17.60°)
was smaller than 10% of the mean angular ranges ( em>200°) of the finger movement dur-
ing the bradykinesia task.
As there were some bad fits in the results of the EM tracking system during hand grasping
movements, another tracking system, such as an optical tracking system, can be used as the
reference system in the future.
In order to improve the accuracy of the bradykinesia parameters, further signal processing
methods must be carried out. The integration method of angular velocity plays a key role in
the calculation of bradykinesia parameters. Also a modified DCM algorithm can be used in
calculating the hand-grasping angle ranges.
7.3 Verification of Analytical Methods for Rigidity Assessment
7.3.1 Impact of Eccentric Application of Force on Force Sensor Box Outputs
Motivation
FSRs have a nonlinear characteristic. Therefore, in the situation that force is introduced out-
side the middle of the FSR’s active area, the result is a lower output than when force is direct-
ly applied to the middle of the active area.
Similarly, when the applied force is performed more towards the edges of the force sensor
box, the result is a lower output than when force is applied to the middle of the box. However,
the magnitude of this effect remains unclear since the characteristics vary among FSRs and
also depend on the chosen contact pad setup.
An experiment was carried out to obtain the deviation between the calculated force value and
its real value (weight value) when an examiner pressed different parts of the force sensor box-
es.
Hypothesis
Experiments and Discussions
87
According to the datasheet of FSR149NS, its accuracy ranges from 5% to 25%. Therefore, the
following requirements should be met:
Accuracy (relative deviation): Val < 25%.
Materials
A series of iron weight plates (2503000 grams)
A rigidity assessment system (included a rigidity cuff and a USB cable)
A laptop with the application software of the rigidity assessment (Visual C# 2010-
based GUI for rigidity assessment and MATLAB R2008b for data analysis)
Parameters
The difference between the measured value and the attached weight load was defined as: Val
Methods
The relative deviations were calculated according to the following formula:
– , (7-7)
where F(out) was the measured force value and F(weight) was the weight value attached to
the force sensor box. The unit of F(weight) was also gram.
The mean absolute value of the relative deviations at a certain point or for a certain weight
was expressed as:
, (7-8)
where N was the number of values at a certain point or for a certain weight in different points.
Experiment Setup
Figure 7-10: Plot with depressed points on force sensor box 1, in which X0 is the central point of the press-
ing area. These same points were marked on the force sensor box 2 as well (© TUM-MIMED, 2012).
As Figure 7-10 shows, four FSR sensors were located underneath the points marked with X11,
X22, X33, and X44. Points X1, X2, X3, and X4 were located at the midways between the center
X0 and X11, X22, X33, and X44, respectively.
Similarly to the calibration of force sensor boxes in Chapter 7.1.2, a series of weight loads
were applied to the force sensor boxes. However, in this experiment, the force was not only
applied to the central point of the force sensor boxes but also to the edges. Ten weight loads
ranging from 250 grams to 3000 grams were exerted on the marked points, shown in Figure
Experiments and Discussions
88
7-10, of each force sensor box. As a result, 90 voltage values for each force sensor box were
acquired.
Derivation Values of the Measurements in Different Points
Table 7-1: Relative deviations of force sensor box 1 and the mean absolute values of the relative devi-
ations. The applied points with the biggest relative deviation are highlighted.
Force
[N]
Relative deviation
X1 X11 X2 X22 X3 X33 X4 X44 X0
2.47 14% 49% 39% 47% 14% -3% -38% -59% 8% 30%
3.45 32% 35% 29% 22% 8% -6% -35% -70% -1% 27%
6.29 19% 29% 17% 19% 7% -15% -39% -74% -1% 24%
5.10 24% 34% 23% 9% 8% -10% -45% -72% 12% 26%
9.90 13% 15% 19% -11% 1% -24% -33% -73% 7% 22%
14.74 6% 8% 9% -5% -16% -37% -41% -73% 5% 22%
19.68 0% -9% 3% -9% -23% -41% -32% -77% 5% 22%
24.57 -10% -12% 1% -19% -31% -53% -33% -53% -2% 24%
38.55 -17% -37% -7% -34% -45% -48% -40% -62% -11% 33%
30.05 -11% -22% -3% -10% -34% -43% -36% -59% -12% 26%
15% 25% 15% 19% 19% 28% 37% 67% 6%
Table 7-2: Relative deviations of force sensor box 2 and the mean absolute values of the relative de-
viations. The applied points with the biggest relative deviation are highlighted.
Force
[N]
Relative deviation
X1 X11 X2 X22 X3 X33 X4 X44 X0
1.49 15% -6% 8% -5% -1% -16% 2% -11% 5% 8%
2.47 4% -10% 5% -10% -6% -24% 17% -14% 13% 11%
3.45 5% -14% 8% 4% -4% -12% 10% -17% 13% 10%
6.29 6% -11% 6% 2% -6% -9% 6% -32% 7% 9%
5.10 3% -12% 9% 15% 6% -6% 11% -14% 8% 9%
9.90 2% 2% -1% -17% -13% -17% -1% -19% 14% 10%
14.74 3% -1% -10% -9% -13% -25% -11% -8% 8% 10%
19.68 4% 1% -14% -14% -29% -33% -19% -41% 2% 17%
24.57 -3% -14% -23% -28% -25% -41% -31% -42% 1% 23%
5% 8% 9% 12% 11% 20% 12% 22% 8%
In the first step, the recorded sensor output was normalized to values of Gramm using Equa-
Experiments and Discussions
89
tion 6-1. Table 7-1 and Table 7-2 show the relative deviations which were calculated accord-
ing to the Equations 7-7 and 7-8.
When the weights, ranging from 2.47 to 30.05 N, were applied to the center of both force sen-
sor boxes ten times, the average deviations of two force sensor boxes were 6% and 8%, re-
spectively.
Summary
When force was applied to the center of the force sensor boxes, the relative deviations were
moderate ( <15%). However, if the force was applied to the edges of the force sensor
boxes, two effects, which impaired the output, could be observed. Firstly, values measured
when applying force more to the edges of the sensor box tended to be lower than when force
was applied to the center. This was likely due to the nonlinear characteristics of the sensors.
Secondly, when force was applied to the edges, the underlying sensor beared a greater share
of the total force and was already closer to saturation than when it was during centric applica-
tion of force.
However, this effect seems to be influenced by the differences in sensitivity of FSRs. Even
though the FSRs had been selected to match in terms of characteristics, deviations depended
strongly on which of the force sensor was exposed to the greatest share of total load. As it can
be seen in column X44 of Table 7-1 (highlighted), the FSR 4 in force sensor box 1 had a very
low sensitivity compared to the other FSRs and therefore influenced the results.
In order to avoid such problem, it is necessary to either select the FSRs more carefully or
connect each FSR in the force sensor boxes to a single analog input (ADC) pin of the micro-
controller and perform individual calibration of each single FSR in the two force sensor
boxes.
In addition, the weight values ranged only from 2.47 to 30.05 N. A larger weight range
(3080 N) should be verified in the future research.
7.3.2 Verification of Wrist Angle Calculation
Motivation
As the torque of elbow passive movement is based on wrist angular displacement, the verifi-
cation of the elbow angle calculation is introduced in this section. The examiner can hold the
patient’s elbow to perform straight movement along a single axis, and then only the move-
ment around the EM generator’s Z-axis was verified in this experiment.
Hypothesis
The calculation of the elbow’s angular displacement during passive movement was based on
the sensor fusion (see Chapter 5.3.2). By comparing with those from the EM system, the an-
gular displacements of the elbow movement measured by the glove monitoring system should
meet the following requirement:
Mean difference of the angular displacements of the two systems during all tested ri-
gidity tasks (elbow movement)
α mz < 0.1 · emz.
Experiments and Discussions
90
Here α emz is the mean angular range of the ten-second elbow movement (peak-to-peak value
in Z-axis) measured by the EM tracking system.
Materials
An NDI Aurora®
EM tracking system with a six-DOF sensor (Northern Digital Inc.,
Canada)
A rigidity assessment system (included a rigidity cuff and a USB cable)
A computer installed with MATLAB R2008b (Mathworks Inc., USA), EM application
software (Northern Digital Inc., Canada), and rigidity assessment software (Visual C#
2010, Microsoft Inc., USA)
Parameters
Maximum difference of the calculated elbow angles from both the EM tracking sys-
tem and the rigidity assessment system during all the ten-second rigidity assessment
tasks
Mean difference of the calculated elbow angles from both the EM tracking system and
the rigidity assessment system during all the ten-second rigidity assessment tasks
Standard deviation of the differences of the calculated elbow angles from both the EM
tracking system and the rigidity assessment system during all the ten-second rigidity
assessment task
Methods
Z-axis elbow angle at a sampled point of the EM tracking system was , which was ac-
quired directly from the NDI EM tracking system. At the same time, the Z axis elbow angle
measured by the rigidity assessment system was , which was calculated using the six-axis
DCM algorithm based on the IMU signals (see Chapter 5.3.2).
Therefore, the difference of angular displacements in a sampled point was:
. (7-9)
The mean difference of the angular displacements during all the ten-second elbow movements
was:
α mz , (7-10)
where N was the number of all sampled points. For one time rigidity assessment, the sampled
points were 1000. There were five valid measurements in this experiment. Thus N was 5000.
The calculation of emz is the same as the method described in Chapter 5.3.2.
Setup
Only one test subject (age: 27 years) was involved in this experiment. He performed the rigid-
ity task six times. As shown in Figure 7-11, a six-DOF EM sensor attached to the wrist was
used in the elbow-motion-tracking experiment. With the help of an examiner, the subject per-
Experiments and Discussions
91
formed elbow flexion and extension movements around the Z axis of the EM generator, while
the end of the elbow acted as a fulcrum.
Figure 7-11: Setup for the verification of elbow angle calculation. The subject performed the same elbow
movement (around the Z-axis) during the rigidity task. An EM sensor was attached to the wrist (Based on
Dai et al., 2013a).
The elbow’s angle values, which were obtained from the EM system directly, were compared
to the elbow angles calculated using the DCM method with the IMU data.
Results
Figure 7-12 shows the angular displacements of elbow’s movement both from the IMU based
rigidity assessment system (DCM fusion algorithm) and from the EM tracking system in the
second measurement. Their difference and the maximum difference over a ten-second task are
also shown in Figure 7-12.
Figure 7-12: Plot of DCM fusion (IMU tracking) versus EM tracking. The elbow movement during a ri-
gidity assessment task was measured with both IMU and EM sensors. IMU signals were processed using
the DCM algorithm. The EM tracking values (angles) were directly obtained from the EM tracking sys-
tem (Based on Dai et al., 2013a).
In the sensor data of the sixth measurement, too many sampling points from the EM data were
missing because of an error in the NDI toolbox; thus only five measurements were analyzed.
Table 7-3 shows the absolute differences between the EM tracking system and the rigidity
assessment system from all the five measurements.
Rigidity cuff
EM sensors
Experiments and Discussions
92
Table 7-3: Absolute differences between the EM tracking system and the IMU based rigidity assess-
ment system
Test Number 1 2 3 4 5
Max. Difference 9.41° 8.27° 12.50° 9.34° 9.90°
Mean Difference 4.40° 4.90° 4.86° 4.70° 4.87°
SD Difference 1.69° 1.82° 2.01° 1.96° 2.05°
The maximum, mean value, and standard deviation of the differences between these two sys-
tems in all test subjects were 12.50°, -4.74°, and 1.91° respectively. The angular ranges of
elbow movements were about 80° during this experiment. Thus the mean difference ( α mz =
-4.74±1.91°) was smaller than 10% of the elbow angular range ( emz >80°).
7.3.3 Verification of Mechanical Impedance Components
Motivation
In order to acquire the correlation between the UPDRS ratings and mechanical impedance,
first the correlation of elastic stiffness and viscosity with the rigidity severity should be inves-
tigated. In the presented experiment, the rigidity assessment algorithm was verified.
Hypothesis
For Cohen's d, an effect size of 0.8 to infinity denotes a “large” effect (Nakagawa et al.,
2007). Therefore, the following requirements should be met:
Effect size d > 0.8.
p-value < 0.05.
Materials
A rigidity assessment system which includes a rigidity cuff and a USB cable
A computer installed with MATLAB R2008b and rigidity assessment software (Visual
C# 2010)
A wooden folding ruler in hard wood (scale range: 1 m)
Parameters and Methods
To assess if the proposed system was able to detect an increase in mechanical rigidity at a
significant level, the four sample reference data were t-tested against the four sample data for
every test subject. Hence a one-side paired t-test between reference state (relaxed) and imitat-
ed rigid state of each volunteer was performed. The difference between the elasticity or vis-
cosity between the relaxed state and imitated rigidity state was described as:
, (7-11)
where
Experiments and Discussions
93
Y=Elasticity (or viscosity) while the test subjects contracted arm muscles;
Z=Elasticity (or viscosity) while the test subjects kept muscles relaxed.
The calculations of elasticity and viscosity are described in Chapter 5.3.3.
The standard deviation, t-test formula, and effect size of X are described as Equations 7-12, 7-
13, and 7-14, respectively.
Standard deviation of X:
N
i
i XXN
XS1
)(1
; (7-12)
One-side paired t-test: NXS
Xt ; (7-13)
Cohen’s XSXd ; (7-14)
where S(x) was the standard deviation of X, N was the number of measurements, and was
the average value of X.
Experimental Setup
As Figure 7-13 shows, nine healthy volunteers (average age: 24.4±4.2 years) were tested with
the rigidity assessment system, yielding eight measurements each. During the first four refer-
ence movements, the volunteers were asked to relax (i.e., with no rigidity), while in the next
four movements they were asked to imitate rigidity.
Figure 7-13: Experiment with the rigidity assessment system (Based on Dai et al., 2013b).
Table 7-4: Ranges and frequencies of four times elbow movements for a test subject during the rigidity
assessment tasks
Number Range Frequency
1 60° 0.5 Hz
2 60° 1.0 Hz
3 120° 0.5 Hz
4 120° 1.0 Hz
Experiments and Discussions
94
Also, during the assessments, the examiner tried to perform different elbow movement ranges
and frequencies according to Table 7-4.
Results
Figure 7-14 shows the torque-displacement plots both in normal condition (i.e., relaxed state)
and imitated rigidity condition.
a) b)
Figure 7-14: Torque-displacement plots (Taken from Dai et al., 2013b). a) normal condition (relaxed),
elbow movement range: 100°, dominant frequency: 1.2 Hz; b) imitated rigidity condition, elbow move-
ment range: 120°, dominant frequency: 1.1 Hz.
The viscosity values, elasticity values, and the frequencies during this experiment were calcu-
lated with the algorithms stated in Chapter 5.3.3.
The viscosity and elasticity were converted to absolute values before calculation. The average
value of viscous modulus with no rigidity (i.e., relaxed state) was 0.26±0.08 N·m/degree, and
0.78±0.45 N·m/degree in the imitated rigidity state. The average value of elastic modulus
with no rigidity (i.e., relaxed state) was 0.99±0.53 N·m/degree, and 3.78±2.85 N·m/degree in
the imitated rigidity state.
Table 7-5: Evaluation of the viscosity. The insignificant value (p>0.05) is highlighted with yellow.
Subject 1 2 3 4 5 6 7 8 9 Mean
value
7.1 12.3 5.4 7.8 12.1 8.1 3.1 4.0 4.8 4.81
S(X) 3.7 6.3 1.45 3.7 2.8 3.6 2.9 1.2 1.4 1.35
Effect
size (d) 1.91 1.94 3.74 2.13 4.40 2.26 1.07 1.35 3.55 1.65
p-value 0.015 0.015 0.002 0.012 0.002 0.01 0.061 0.036 0.003 <<
0.001
Table 7-5 and Table 7-6 show the viscosity and elasticity of the subjects assessed by the rigid-
ity assessment system, respectively.
Experiments and Discussions
95
Table 7-6: Evaluation of the elasticity. The insignificant value (p>0.05) is highlighted with yellow,
while the negative elasticity is highlighted with green.
Subject 1 2 3 4 5 6 7 8 9 Mean
value
0.94 1.00 0.54 0.80 2.39 0.14 0.58 -0.26 0.65 0.75
S(X) 0.44 0.21 0.27 0.12 0.71 0.13 0.23 0.12 0.13 0.75
Effect
size (d) 2.11 4.78 1.99 6.77 3.34 1.04 2.43 -2.20 4.81 1.00
p-value 0.012 0.001 0.014 <0.001 0.003 0.065 0.008 - 0.001 <<0.001
For eight of the nine test subjects, the proposed system was able to detect a significant change
in viscosity (p<0.05). For eight of the nine test subjects, a significant change of the elbow
joint elasticity (p<0.05) was also detected. The effect size (Cohen's d) of viscosity and elastic-
ity between normal state and imitated rigidity were therefore, “large”, at 1.61 and 1.36
(d>0.8), respectively.
Influence of Angular Range and Frequency of the Elbow Movement
The viscosity and elasticity versus elbow ranges and frequencies are shown in Figure 7-15 to
Figure 7-18, which were produced using the freely available statistical software package R
(version 2.15.2).
Figure 7-15: Plot of viscosity against the range of passive elbow movement (© TUM-MIMED, 2012).
Angular range [º]
Vis
co
sity [
N·m
·s/º
]
Measured value while the test
performed imitated rigidity
Measured value while the test
was in relaxed state
Experiments and Discussions
96
Figure 7-16: Plot of elasticity versus the range of passive elbow movement (© TUM-MIMED, 2012).
Figure 7-17: Plot of viscosity over the frequency of passive elbow movement (© TUM-MIMED, 2012).
Frequency [Hz]
Ela
sticity
[N·m
/º]
Angular range [º]
Vis
co
sity [
N·m
·s/º
]
Measured value while the test
performed imitated rigidity
Measured value while the test
was in relaxed state
Measured value while the test
performed imitated rigidity
Measured value while the test
was in relaxed state
Experiments and Discussions
97
Figure 7-18: Plot of elasticity over the frequency of passive elbow movement (© TUM-MIMED, 2012).
The result shows that the frequencies and ranges of these passive elbow movements were not
exactly the same as the settings in Table 7-7, because it was very difficult for the examiner to
keep accurate movement frequency or range in an assessment task. The mean and SD values
of viscous modulus and elastic modulus in different movement ranges and frequencies are
displayed in Table 7-7.
Table 7-7: Mean and standard deviation of the absolute viscosity and elasticity
Range Frequency
60° 120° 0.5 Hz 1 Hz
Viscosity
(normal) 0.28±0.08 0.23±0.09 0.25±0.08 0.26±0.09
Viscosity
(rigid) 0.79±0.46 0.77±0.44 0.80±0.55 0.76±0.39
Elasticity
(normal) 1.15±0.83 0.82±0.51 0.81±0.51 1.18±0.62
Elasticity
(rigid) 3.55±3.43 4.01±2.47 3.55±2.95 4.02±2.96
The result shows that the frequency and range of elbow movement had little effect on the vis-
cosity. In contrast, movement frequency had a greater effect on the elasticity, which might
have negative influences when the examiner flexed and extended the forearm at different
speeds. As a result, if the neurologist wants to obtain the mechanical impedance according to
the formula of mechanical impedance (see Chapter 5.3.3), it is important to keep the same
frequency. Compared with normal state (relaxed), elasticity varies largely in the imitated ri-
gidity state.
Frequency [Hz]
Ela
sticity [
N·m
/º]
Measured value while the test
performed imitated rigidity
Measured value while the test
was in relaxed state
Experiments and Discussions
98
According to the formula of mechanical impedance, elasticity depends on the movement fre-
quency and range, and other factors. Thus, the mean value of elastic modulus had a large
standard deviation. Another reason is that the test subjects did not perform imitated rigidity in
the same state, which means the imitated rigidity varied.
7.4 Experiment of Tremor Assessment
Clinical experiments of patients with tremors were carried out with the glove monitoring sys-
tem. The glove monitoring system in this section refers to the tremor/bradykinesia assessment
system (prototype for tremor and bradykinesia assessments).
Hypothesis
The tremor amplitude correlation between the glove monitoring system and the clinical rat-
ings should meet the requirement:
Correlation coefficient r > 0.84 (Giuffrida et al., 2009).
Materials
A tremor/bradykinesia assessment system (included a command module with a sensor
board, and a JTAG USB cable)
A computer installed with MATLAB R2008b (MathWorks Inc., USA) and the GUI
(LabVIEW 2010 Evaluation Version, National Instruments Corp., USA) of the
tremor/bradykinesia assessment system.
Parameters and Evaluation Methods
As described in Section 5.3.1, the tremor amplitude calculated by the glove monitoring sys-
tem was R, while the tremor amplitude judged by the surgeon was D. Then the correlation
coefficient between these two parameters was:
, (7-15)
where N is the amount of assessed patients.
For the valid state detection (see Chapter 5.3.1), only the gyroscope signal was utilized. Pa-
rameters for the valid state detection both in the frequency domain and time domain were:
; (7-16)
, (7-17)
where was the matrix of peak-to-peak values of the triple-axis gyroscope signals during
a ten-second tremor task, peak power was the power estimation around the dominant tremor
frequency (±0.3 Hz).
Experiment Setup
A total of five patients with tremors were tested with this wearable system. The tremor severi-
Experiments and Discussions
99
ty of these patients ranged from 1 to 3 (UPDRS tremor scores: D). The measurements in these
patients were performed after stopping their medication for more than 24 hours.
Rest, postural, and action tremor assessment tasks were performed. However, for each patient,
only the hand side with a more severe tremor was assessed. Each tremor assessment task last-
ed for ten seconds.
The tremor amplitude was represented the tremor severity in this system. Time-frequency
analyses and statistical analyses were carried out based on the IMU signals. The valid state
detection algorithm was performed for each assessment task.
As shown in Figure 7-19, a PD patient performed three tremor assessment tasks in turn ac-
cording to the instructions of a surgeon.
a) b) c)
Figure 7-19: Experiments with the tremor/bradykinesia assessment system. a) rest tremor; b) postural
tremor; c) action tremor.
Each patient performed 3–5 times each task repeatedly. Each task lasted ten seconds.
Results
Figure 7-20 shows the IMU signals of the glove monitoring system and their power spectra
for a PD patient with slight rest tremor (D=1). In Figure 7-20 (a), the tremor occurred only for
several seconds and with varying amplitude. In Figure 7-20 (b), there was only a small ampli-
tude fluctuation during the ten-second period.
The tremor amplitude calculation with PSD method depends on the precondition that the
tremor movement is stable and the power estimation around the dominant frequency is sharp.
Therefore, the signals in Figure 7-20 (b) were able to be used to calculate the tremor ampli-
tude. The assessment tasks which had bigger tremor amplitude fluctuation than Figure 7-20
(b) were discarded during signal processing because the tremors were not in a stable condi-
tion. Although Heldman et al. (2011b) indicated that the tremor amplitude calculated with
PSD correlated well with clinical ratings even for a broad power spectra distribution, the
power distribution in other frequency points still influenced the tremor amplitude calculation
in these measurements.
As shown in Figure 7-20, (d) is the normal situation in frequency domain while (c) is not
good for tremor calculation with the PSD method (Timmer et al., 1997).
For the action tremor assessment, it was difficult to separate the active movement with tremor
even with different band-pass filters. Its power spectrum included multiple frequency compo-
nents. A better signal filter should be adopted in the future (Timmer, et al. 1993).
Experiments and Discussions
100
When performing the signal processing algorithms, the rest and postural tremor tasks were
assessed as valid state or invalid state for PSD estimation. The amplitude of action tremor was
not included in the present results.
a) invalid state in time domain (Vt = 0.64);
b) valid state in time domain (Vt = 0.23);
c) invalid state in frequency domain (Vf = 69.4%); d) valid state in frequency domain (Vf = 91.3%).
Figure 7-20: Tremor state of a PD patient (UPDRS tremor score: D=1). a) and b): waveforms of triple-axis
raw accelerations and triple-axis angular velocities (rest tremor); c) and d): combined power spectra of
the triple-axis acceleration signals (bottom chart) and the combined power spectra of the angular velocity
signals (upper chart). b) and d) are for the same tremor assessment task.
In order to investigate the correlation between the inertial sensor outputs and clinical ratings,
the outputs of the gyroscope and the accelerometer were analyzed independently as well. As
shown in Table 7-8, the parameters, calculated with the sensor signals, were compared to the
judgments of a surgeon according to the UPDRS and TETRAS ratings. These results were
from the average values of valid rest and postural tremor tasks.
Table 7-8 shows that the gyroscope signal had a stronger correlation (r=0.92) with the clinical
scores than the accelerometer signal (r =0.81).
Both the RMS value of the gyroscope signal and peak power from all IMU outputs have the
biggest correlation with the clinical scores (r=0.92). In the program of the glove monitoring
system, the frequency range of peak power was 0.6 Hz. Thus, the frequency range should be
wider in the next version of this glove monitoring system.
Frequency (Hz) Frequency (Hz)
Experiments and Discussions
101
Table 7-8: Results for rest and postural tremor assessments. Here acc. represents accelerometer; gyro.
means gyroscope; RMS means root-mean-square; ln means natural logarithm; R is the predicated
tremor score from the glove monitoring system; r denotes the correlation coefficient between the rela-
tive parameters calculated by the IMU signals and clinical scores judged by the surgeon.
It can also be seen from Table 7-8 that the proportion of rotational movement and translation
movement of a tremor [ln(gyro. Power) versus ln(acc. Power)] varies from patient to patient.
More tremor measurements are needed to acquire further symptom information and better
regression coefficients.
Figure 7-21 shows the dominant tremor frequencies of a patient during three times rest tremor
tasks (numbers 1 to 3) and three times postural tremor tasks (numbers 4 to 6). The difference
of dominant frequencies from gyroscope and accelerometer was small (<0.2 Hz). For all the
measured patients, the dominant tremor frequency was constant or with small fluctuations
even if the tremor amplitude changed during ten-second task period.
Figure 7-21: Dominant tremor frequency of a PD patient (UPDRS tremor score D=1) during the assess-
ment tasks of rest tremor (numbers 1 to 3) and postural tremor (numbers 4 to 6). The dominant tremor
frequencies were calculated from accelerometer signals and gyroscope signals, respectively.
Discussion
Results indicate that the tremor frequency was stable for all patients; however, the tremor am-
plitude fluctuated all the time. The quantitative assessment of tremor, with an IMU and adap-
tive algorithms, provided an objective rating to classify rest or postural tremor severity even
with a small scale.
Experiments and Discussions
102
7.5 Experiment of Bradykinesia Assessment
Motivation
For the measurements in a healthy subject or a patient with mild bradykinesia, the signals
obtained from the gyroscope have a consistent amplitude and frequency and appear sinusoi-
dal. However, the signals from a patient with severe bradykinesia have a much lower and in-
consistent amplitude and frequency.
Thus the mean and standard deviation of amplitude and frequency during bradykinesia meas-
urement tasks represent the parameters of bradykinesia.
The prototype should be tested both with healthy volunteers and PD patients to demonstrate a
difference between the PD patients and healthy subjects. The glove monitoring system in this
section refers to the tremor/bradykinesia assessment system.
Hypothesis
The correlations of bradykinesia parameters between the glove monitoring system and clinical
ratings should meet the following requirement:
Correlation coefficient r > 0.79 (Giuffrida et al., 2009).
Parameters and Evaluation Methods
As described in Section 5.3.2, the bradykinesia parameter calculated by the glove monitoring
system is R, while the severity of bradykinesia judged by the surgeon is D. Then the correla-
tion coefficient between these two parameters is
, (7-18)
where N is the amount of assessed patients.
Along with dominant frequency of hand grasping, mean and SD values of hand grasping
ranges ( ) were regarded as bradykinesia parameters. In addition, we defined the product of
dominant frequency and mean range as the modified mean range.
Materials
A tremor/bradykinesia assessment system which included a JTAG USB cable and a
command module with a sensor board
A computer installed with the GUI of the tremor/bradykinesia assessment system
(LabVIEW 2010 Evaluation Version, National Instruments Corp., USA) and MAT-
LAB R2008b (Mathworks Inc., USA)
Experiment Setup
Three healthy volunteers (average age: 49.0±16.6 years) and five PD patients (average age:
75.5±11.1 years) should execute hand grasping movement as quickly and widely as possible
for ten seconds, and repeat the assessment task three times for each subject. However, a se-
vere PD patient (UPDRS bradykinesia score D=4) was unable to perform the bradykinesia
Experiments and Discussions
103
task.
One time grasping movement included several grasp cycles, which was the number of peak-
to-peak ranges.
An IMU, which was attached to the middle finger, was used to measure the angular displace-
ment of the middle finger movement during the bradykinesia assessment task. The dominant
grasping frequency, mean, and standard deviation of hand grasping ranges were used as the
severity features of bradykinesia.
The bradykinesia parameters, calculated by the glove monitoring system, were compared to
the ratings of a surgeon.
Results and Discussion
Figure 7-22 shows three ten-second waveforms of hand grasps and their PSD figures. For
some patients, as shown in Figure 7-22 (b), action tremor appeared during the bradykinesia
assessment task.
Figure 7-22: Ten-second hand grasps and their PSD figures. a) healthy subject, age: 72 years; b) PD pa-
tient with tremor and bradykinesia, age: 82 years, UPDRS bradykinesia score D= 1; c) PD patient, age: 86
years, UPDRS bradykinesia score D= 3. The peak powers in the figure were calculated on a weighted
scale.
In addition, as shown in Figure 7-22 (c), there was a delay time for a patient with severe
bradykinesia to start the hand grasping movement after the instruction from the examiner.
This situation is the symptom of akinesia (difficulty initiating movement).
The three healthy subjects had a dominant frequency of 1.16 Hz and a standard deviation of
0.11 Hz. The parameter values are presented in Table 7-9. Table 7-9 shows that the patients
with severe bradykinesia executed grasping movements with lower dominant frequencies.
As Table 7-9 shows, the dominant frequency of hand grasps and the modified mean range had
the higher correlations with the UPDRS score (r=-0.94 and -0.86 respectively).
0 2 4 6 8 10
0
100
-100
0
200
-200
0
150
-100
Time (s)
0 2 4 6 8 10
0 2 4 6 8 10 0 2 4 6
0 2 4 6
0 5 10
Frequency (Hz)
5
10
0
2
4
0
2
4
0
Angle
( )
Angle
( )
Angle
( )
Peak p
ow
er
Peak p
ow
er
Peak p
ow
er
a)
b)
c)
Combined angle of hand grasps ( )α PSD figure of grasping angles
Experiments and Discussions
104
Table 7-9: Results of bradykinesia assessment tasks. Freq. represents frequency; r denotes the correla-
tion coefficient between the relative parameters and clinical scores.
Subjects Dominant Freq. Mean Range SD Ranges Modified
Mean Range
UPDRS
Score (D)
Healthy 1 1.09 Hz 202.3° 10.6° 220.5 Hz·° 0
Healthy 2 1.29 Hz 255.4° 11.5° 329.5 Hz·° 0
Healthy 3 1.10 Hz 289.8° 19.8° 318.8 Hz·° 0
Patient 1 0.88 Hz 241.6° 10.2° 212.6 Hz·° 1
Patient 2 0.60 Hz 181.7° 9.47° 109.0 Hz·° 1
Patient 3 0.44 Hz 212.8° 21.0° 93.6 Hz·° 2
Patient 4 0.32 Hz 221.8° 3.6° 71.0 Hz·° 3
r -0.94 -0.36 -0.32 -0.86
There were also subjects, such as Healthy 1 and Patient 1, whose mean grasping range deviat-
ed from the normal situation. The reason may be the age of subjects. The healthy subject was
too old to reach great amplitudes during hand grasping movements. On the other hand, the PD
patient with a smaller age was able to grasp at a higher amplitude. Therefore, the obtained
parameters need to be modified according to the subject’s age. For the SD values of the grasp-
ing ranges, there was only a small correlation coefficient (r=-0.32) to the UPDRS ratings in
this experiment.
As a conclusion, the difference of the parameters between the healthy subjects and PD pa-
tients was significant. Further clinical measurements should be performed in the future.
7.6 Conclusion
As a force sensor box of the rigidity assessment system had four FSRs and the force-
resistance characteristic was nonlinear, a two-term Gaussian regression function was used for
each force sensor box calibration.
A better cognition of the analytical methods in this system was achieved with these verifica-
tion results, providing the necessary science and engineering to guide future signal processing
methods. Then the clinical experiments could be performed.
The verification experiment of tremor amplitude and frequency indicates that the dominant
tremor frequencies from both systems had little difference (0.115±0.144 Hz). For the imitated
tremor task with consistent movement, the calculated tremor amplitude using the IMU based
system correlated well with the tremor amplitude from the EM system (r=0.97). The results of
clinical experiment indicate that the tremor amplitude calculation with the PSD method was
valid only in the constant state. Thus the tremor tasks were divided into valid state and invalid
state. Rest and postural tremors in valid state were correlated with the UPDRS scores
(r=0.92). For the action tremor amplitude calculation, quadratic mean method or other algo-
rithms should be employed. The dominant tremor frequency of a patient for one type of trem-
or task had small fluctuations even in an invalid tremor state (r=0.996) (Elble et al., 2006).
Experiments and Discussions
105
Further measurements are needed to modify the regression coefficients in the regression func-
tion for tremor calculation.
The difference between parkinsonian tremor and ET needs to be investigated.
The analysis as to whether there are single or multiple dominant frequencies during tremor
assessment tasks also needs to be performed (Mostile et al., 2010).
The clinical experiment indicates that the components of rotational and linear movements of
different subjects’ tremors were different.
According to the results of clinical measurements, the dominant frequency and modified
mean range of hand grasps correlated well with the UPDRS ratings. Their correlation coeffi-
cients were 0.94 and 0.86 respectively.
As quantitative rigidity assessment in PD is difficult and depends on many factors, a compari-
son experiment was carried out. Elastic and viscous values will be obtained through a least
squares estimation with all the data. Nine healthy subjects were tested with this system in two
experimental conditions: 1) normal state (relaxed); and 2) imitated rigidity state. In addition,
the subjects performed the assessment task with different frequencies and elbow movement
ranges. The result indicates that viscosity and elasticity in the imitate rigidity condition are
bigger than the normal condition (relaxed state). The effect sizes (Cohen's d) of the viscosity
and elasticity between normal state and imitated state are 1.61 and 1.36 respectively, which
means the difference is significant. Thus, this system can detect the ON-OFF fluctuations of
parkinsonian rigidity. Both the wrist movement angle and frequency have a small effect on
the viscosity, but have an elevated effect on the elasticity.
The future research consists of carrying out measurements with more PD patients. With the
measurement data, the correlation between mechanical impedance (viscosity, elasticity, and
movement frequency) and UPDRS scores can be determined. After that, the system can be
used for rigidity assessment in PD.
Nevertheless, attention must be paid to the signal processing methods:
Influence of noise in tremor assessment when the PD patient has a slight tremor.
Inconsistent movement and its influence in tremor amplitude during the tremor as-
sessment task, especially for peak power calculation with the PSD estimation.
Combination of the six-axis signals in both a gyroscope and an accelerometer, which
represent the rotational and the linear movement respectively.
Correlation between different sensor signals and the UPDRS ratings.
Repeatability of parameters with the same patient at different times.
Accurate quantification action tremor.
Summary and Outlook
106
8. Conclusions and Outlook
8.1 Conclusions
A portable monitoring system used to quantify all primary neurological symptoms in the PD
and ET during the DBS surgery is presented in this thesis.
According to the requirements of neurosurgeons, the concept of a glove monitoring system
used in DBS surgery is presented. This system is focused on the symptom assessments of PD
and ET.
According to the definitions of UPDRS and TETRAS scores and the situation during DBS
surgery, several of such tasks were chosen to detect the stimulating effects of the electrical
stimulation. Each assessment lasts for ten seconds. Passive flexion and extension of the elbow
is used to assess rigidity in this study. Tremor, bradykinesia, and rigidity assessments are per-
formed in a system based on IMUs and FSRs. An IMU can be used to measure both the rota-
tional and linear movement, using the combination of a triple-axis accelerometer and a triple-
axis gyroscope. This system consists of a glove part and a computer part. The communication
between these two parts is based on a USB cable. The measured parameters in different elec-
trode placements and stimulating intensities are listed in the GUI for comparison.
The positions of sensors are the wrist and fingers. All the sensors have no contact with the
human body.
The first version of the glove monitoring system based on the above concept includes two
portable prototypes: one for tremor/bradykinesia assessment and the other for rigidity assess-
ment. A wearable textile glove, two inertial measurement units (IMUs), two force sensor box-
es, a computer, and other components were included in the prototypes. The prototype of the
tremor/bradykinesia assessment system was based on an IMU attached to a finger. The first
prototype of the rigidity assessment system was based on two force sensor boxes and an IMU
on the wrist.
The verification of the analytical methods of the two prototypes was performed, compared
with an EM system. Because the quantitative rigidity assessment of PD patients is difficult
and depends on many factors, a comparison experiment was carried out. An examiner flexed
and extended the elbow joint of nine volunteers through a rigidity assessment cuff attached
around the wrist, both with the relaxed state and imitated rigidity state. The result indicates
that viscosity and elasticity in the imitated rigidity condition were bigger than they were in
normal conditions (relaxed state).
This system was tested in the hospital but outside the operating room first. Five patients were
tested with the tremor/bradykinesia monitoring system. The tremor measurement results indi-
cated that the tremor amplitude calculation with the PSD method was valid only in the con-
stant state, which was regarded as a valid state. For the action tremor amplitude calculation,
the quadratic mean method or other algorithms should be employed. The dominant tremor
frequency of a patient for one type of tremor task included small fluctuations even in an inva-
lid tremor state. The bradykinesia measurement results indicate that the dominant frequency
and modified mean range of hand grasps correlate well with the severity of bradykinesia.
Summary and Outlook
107
As a result, the prototype for tremor and bradykinesia assessment can quantify tremor and
bradykinesia. However, few patients were measured in this study. More measurements of pa-
tients with PD and ET are needed.
The prototype for rigidity assessment can detect the ON-OFF fluctuations of parkinsonian
rigidity. With further measurement data, the correlation between the UPDRS score and me-
chanical impedance (viscosity, elasticity, and movement frequency) and UPDRS score can be
determined.
Based on the disadvantages of the first version of the glove monitoring system and the sug-
gestions from the surgeons, a combined prototype for the assessment of all three primary
symptoms was also presented.
8.2 Outlook
In this thesis, the glove monitoring system in the DBS application is present. It is the world’s
first objective assessment system for the quantification of three primary symptoms of PD and
ET.
However, there are still some approaches to improve the system in the next step:
1) Further clinical measurements should be performed to modify the regression coeffi-
cients in the algorithms with the combined prototype.
2) Further signal processing algorithms need to be developed based on the hardware and
the measurement data. Automatic calibration algorithms during symptom assessments
can be adopted in addition to the initial calibration procedures.
3) The difference between parkinsonian tremor and ET needs to be further investigated.
4) The analysis as to whether there are single or multiple dominant frequencies during
tremor tasks also needs to be performed (Post et al., 2005).
5) The multi-symptom phenomenon in a single assessment task should be investigated.
6) Nine-axis sensor fusion algorithms and other statistical methods can be incorporated in-
to future designs, especially for the tremor quantification, to get better performance.
Although this system is supposed to be used during DBS surgery, there is no clinical meas-
urement in the operating room yet. Thus, there is still much work to be done in the near fu-
ture:
1) Dyskinesia assessment is important to detect the side effects of DBS surgery. It should
be integrated into the system in the next version.
2) For the comparison of slight symptoms during DBS surgery, much work should be
done. Because the symptoms fluctuate all the time, it is important to measure the stable
parameters.
3) A series of validations and modifications of this system should be carried out according
to the requirements of DBS surgery.
Summary and Outlook
108
4) This system can be combined with the MRI, MER or iMRI systems for the application
of DBS surgery. The effective combination of these different systems will result in a
more complete and accurate severity quantification of the symptoms of PD and ET.
5) Clinical experiments should be performed during DBS surgery.
Only after further experiments and compulsory validations both outside and inside the operat-
ing room is this system supposed to help select the optimal target location and the settings of
the stimulation parameters of the DBS electrodes during and after DBS surgery.
In addition, a glove monitoring system with a wireless communication interface can be im-
plemented for the extended applications, for example, the assessments of symptoms before
and after DBS surgery or at the patients’ home.
Glossary
109
9. Glossary
ADC An analog-to-digital converter converts a continuous voltage to
a digital value that represents the voltage.
AHRS An attitude heading reference system, which are either solid-
state or MEMS gyroscopes, accelerometers, and magnetometers
on all three axes, provide heading, attitude, and yaw infor-
mation. A form of nonlinear estimation, such as a Kalman filter,
is typically used to compute the solution from these multiple
sources.
CE Consumer electronics are electronic devices intended for every-
day use. Apple Inc. has invented the consumerization of infor-
mation technology in the 2010s.
CNS The central nervous system, which consists of the brain and
the spinal cord, contains the majority of the nervous system.
DBS Deep-brain stimulation is a surgical procedure used for some
types of disabling neurological symptoms, which includes PD,
ET, dystonia, and chronic pain.
DCM A direction-cosine-matrix (rotation matrix) is a 3x3matrix con-
taining the cosines of the rotation (each of the 9 possible pairs of
axes) of one frame relative to another.
DOF The degree of freedom of a mechanical system is the number of
parameters that may vary independently.
EM The electromagnetic tracking system is a spatial measurement
system which determines the location and orientation of an ob-
ject that is attached to sensor coils.
ET Essential tremor is a common progressive neurological move-
ment disorder and appears in the arms or hands only in the state
of movement such as eating or writing.
FSR Force sensing resistor is a passive component whose resistance
changes when a force or pressure is applied.
GPi Globus pallidus or paleostriatum is a sub-cortical structure of
the brain which performs the regulation of voluntary movement.
GUI The graphical user interface represents the information and in-
teractions available to a user through graphical icons and visual
indicators and as opposed to text-based interfaces.
Glossary
110
iMRI Interventional MRI is a real-time MRI which can be used in
surgery conditions. Asleep iMRI-guided DBS implantation has
many advantages. The electrodes are positioned with the patient
asleep in an MRI scanner instead of awake in the operating
room.
IMU Inertial measurement unit is a single unit that measures a sub-
ject’s velocity, orientation, and gravitational forces in three di-
mensions, using a combination of accelerometers and gyro-
scopes. It is the main component of an inertial navigation sys-
tem.
LID Levodopa-induced dyskinesia is a form of dyskinesia associated
with anti-parkinsonian medications (Levodopa). It often in-
volves the presence of involuntary movement (chorea, dystonia,
and athetosis) and diminished voluntary movements. DBS can
reduce LID or reduce L-DOPA dosage.
MBRS The Modified Bradykinesia Rating Scale rates the bradykinesia
expressions of speed, amplitude, and rhythm separately, while
the UPDRS rates bradykinesia as a whole. Thus more infor-
mation is provided for different PD patients.
MCU A microcontroller (µC, uC or MCU) is a small computer on a
single chip containing a processor core, memory, an I/O control
unit, and a clock.
MEMS Microelectromechanical systems or micro systems technology is
a technology that can be defined as miniaturized mechanical and
electro-mechanical elements (i.e., devices and structures) of
small scale that are made using the techniques of
microfabrication. Their general components are a microproces-
sor and several components that interact with the surroundings
such as microsensors.
MER Microelectrode recording is a neurosurgical technique of single
neuron activity. MER is realized with the patient awake and
interacting by performing a series of motor tasks so that the neu-
rophysiological mapping of the target is enhanced by functional
mapping to localize the optimal target of the DBS electrode.
MF Motor fluctuations (or ON-OFF fluctuations) refer to the state
changes between symptoms controlled (ON state) and symp-
toms not controlled (OFF state, symptoms are poor or not re-
sponding to the medication).
MRI Magnetic resonance imaging is a radiology medical imag-
ing technique used to visualize body structures in detail.
PD Parkinson’s disease is a central nervous system disease that
leads to movement disorders such as tremor, stiffness, and diffi-
Glossary
111
culty with walking and balance. It appears mainly in people
more than 60 years old, and the risk of developing PD goes up
with age.
PSD The power spectral density describes how the power estimation
of a signal or time series is distributed on the frequency domain.
The PSD is the Fourier transform of the autocorrelation function
of the signal which is a wide-sense stationary random process. If
the signal is not stationary, then its autocorrelation function is a
function of two variables. A similar technique, instead of PSD,
may be used to estimate a time-varying spectral density.
QFN The Quad Flat No-Lead is a surface mount plastic package with
leads located on four sides of the bottom package.
RFI The radio frequency interference refers to the information being
transmitted across unshielded copper cable, which can be inter-
fered with by the noise caused by other radio frequencies.
RMS The root mean square (quadratic mean) is a statistical measure
of the magnitude (the average of the squares) of a randomly
varying quantity.
STN The subthalamic nucleus is a small lens-shaped nucleus in the
brain, a component of the basal ganglia control system, and is
involved in action selection.
TETRAS The Essential Tremor Rating Assessment Scale consists of 10
items in which action tremor is rated 0–4 in 0.5 intervals. It ex-
cellently rates ET in term of estimated tremor amplitude.
TWI The two-wire interface protocol, which is the same as the IIC
protocol, uses two wires for serial data transmission between
two or more chips in asynchronous mode.
UPDRS The Unified Parkinson's Disease Rating Scale is a clinical rating
scale used to evaluate the symptom severities of PD by inter-
view and clinical observation. Other rating scales for PD, such
as the Hoehn and Yahr scale and the Schwab and England Ac-
tivities of Daily Living Scale, are included in the revised
UPDRS.
USB The universal serial bus (USB) is an industry standard that de-
fines the cables, connectors, and communications protocols used
in a 4-pin interface for connection, communication, and power
supply between a computer and another electronic device.
Bibliography
112
10. Bibliography
Abramowitz, M & Stegun, I.A. (1964): Handbook of mathematical functions with formulas,
graphs, and mathematical tables, New York: Dover.
Allen, D.P.; Playfer, J.R.; Aly, N.M.; Duffey, P.; Heald, A.; Smith, S.L. & Halliday D.M.
(2007): “On the use of low-cost computer peripherals for the assessment of motor dysfunction
in Parkinson's disease--quantification of bradykinesia using target tracking tasks”, IEEE
Trans. Neural. Syst. Rehabil. Eng., vol. 15, no. 2, pp. 286–294.
Bamberg, S.; Benbasat, A.Y.; Scarborough, D.M.; Krebs, D.E. & Paradiso, J.A. (2008): “Gait
analysis using a shoe-integrated wireless sensor system”, IEEE Trans. Inf. Technol. Biomed.,
vol. 12, no. 4, pp. 413–423.
Bittar, R.G.; Burn S. C.; Bain, P.G.; Owen, S.L.; Joint, C.; Shlugman, D. & Aziz, T.Z. (2005):
“Deep brain stimulation for movement disorders and pain”, Journal of clinical Neuroscience,
vol. 12, no. 4, pp. 457–463.
Bland, J.M. & Altman, D.G. (2003): “Applying the right statistics: analyses of measurement
studies”, Ultrasound Obstet. Gynecol., vol. 22, pp. 85–93.
Bour, L.J.; Contarino, M.F.; Foncke, E.M.; de Bie R.M.; van den Munckhof, P.; Speelman,
J.D. & Schuurman, P.R. (2010): “Long-term experience with intraoperative microrecording
during DBS neurosurgery in STN and GPi.”, Acta Neurochir., vol. 152, pp. 2069–2077.
Burkhard, P.R.; Langston, J.W. & Tetrud, J.W. (2002): “Voluntarily simulated tremor in
normal subjects”, Neurophysiol. Clin., vol. 32, no. 2, pp. 119–126.
Burkhard, P.R.; Shale, H.; Langstom, J.W. & Tetrud, J.W. (1999): “Quantification of dyski-
nesia in Parkinson’s disease: validation of a novel instrumental method”, Mov. Disord., vol.
14, no. 5, pp. 26–27.
Carmeli, E.; Vatine, J.J.; Peleg, S.; Bartur, G. & Elbo, E. (2009): “Upper limb rehabilitation
using augmented feedback: Impairment focused augmented feedback with HandTutor”, Vir-
tual Rehabilitation International Conference, pp. 220.
Charles S.M. (2003): “Parkinson’s Disease: Overview and Current Abstracts”, New York:
Nova Science Pub Inc., pp. 7–9.
Chaudhuri, K.R. & Ondo, W.F. (2011): “Handbook of Movement Disorders”, London:
Springer Healthcare Ltd, pp. 1–2.
Chwaleba, A.; Jakubowski, J. & Kwiatos, K. (2003): “The measuring set and signal pro-
cessing method for the characterization of human hand tremor”, International Conference on
CAD Systems in Microelectronics (CADSM), pp. 149–154.
Crawford, P. & Zimmerman, E.E. (2011): “Differentiation and diagnosis of tremor”, Ameri-
can Family Physician, vol. 83, no. 6, pp. 697–702.
Bibliography
113
Czabke, A.; D’Angelo, L.T.; Niazmand, K.; & Lueth, T.C. (2009): “ Ein kompaktes System
zur Erfassung und Dokumentation von Bewegungsgewohnheiten”, Conference on Ambinent
Assisted Living -AAL, Berlin, pp. 1–5.
Dai, H.; Otten, B.; Mehrkens, J.H.; D’Angelo, L.T. & Lueth, T.C. (2013a): “A novel glove
monitoring system used to quantify neurological symptoms during deep-brain stimulation
surgery”, IEEE Sensors Journal, vol. 13, no. 9, pp. 3193–3202.
Dai, H.; Otten, B.; Mehrkens, J.H. & D’Angelo, L.T. (2013b): “A portable system for quanti-
tative assessment of parkinsonian rigidity”, IEEE EMBC Conf., pp. 6591–6594.
D’Angelo, L.T.; Czabke, A.; Niazmand, K. & Lueth, T.C. (2010): “ART–a new concept for
an activity recorder and transceiver”, IEEE EMBS Conf., Buenos Aires, pp. 2132–2135.
D’Angelo, L.T.; Michael, S.; Paul, N. & Lueth, T.C. (2011): “A sensor network to iPhone
interface separating continuous and sporadic processes in Mobile Telemedicine”, IEEE EMBS
Conf., Boston, pp. 1528–1531.
Deuschl, G.; Bain, P. & Brin, M. (1998): “Consensus statement of the movement disorder
society on tremor”, Movement Disorders, vol. 13, pp. 2–23.
Dipietro, L., Sabatini, A.M. & Dario, P. (2008): “A survey of glove-based systems and their
applications”, IEEE Transactions on Systems, Man and Cybernetics, vol. 4, pp. 461–482.
Edwan, E.; Zhang, J.Y.; Zhou, J.C. & Loffeld, O. (2011): “Reduced DCM based attitude es-
timation using low-cost IMU and magnetometer triad”, 8th
Workshop on Positioning Naviga-
tion and Communication (WPNC), pp. 1–6.
Elble, R.J.; Pullman, S.L.; Matsumoto, J.Y.; Raethjen, J.; Deuschl, G.; Tintner, R. & Tremor
Research Group (2006): “Tremor amplitude is logarithmically related to 4- and 5-point tremor
rating scales”, Brain, vol. 129, no. 10, pp. 2660–2666.
Elble, R.J. & Deuschl, G. (2011): “Milestones in tremor research”, Movement Disorders, vol.
26, no. 6, pp. 1096–1105.
Ellermeier, W. & Faulhammer, G. (2000): “Empirical evaluation of axioms fundamental to
Stevens’s ratio-scaling approach: I. Loudness production”, Perception & Psychophysics, vol.
62, no. 8, pp. 1505–1511.
Espay, A.J.; Giuffrida, J.P.; Chen, R.; Payne, M.; Mazzella, F.; Dunn, E. et al. (2011): “Dif-
ferential response of speed, amplitude, and rhythm to dopaminergic medications in Parkin-
son’s disease”, Movement Disorders, vol. 26, no. 14, pp. 2504–2508.
Endo, R.; Yokoe, M.; Fukawa, K.; Sakoda, S. & Akazawa, K. (2007): “Measurement system
of finger-tapping contact force for quantitative diagnosis of Parkinson’s disease”, IEEE EMBS
Conf., pp. 1354–1357.
Endo, T.; Okuno, R.; Yokoe, M.; Akazawa, K. & Sakoda, S. (2009): “A Novel method for
systematic analysis of rigidity in Parkinson’s disease”, Movement Disorders, vol. 24,
pp. 2218–2224.
Gowers, W.R. (1886): “A Manual of Diseases of the Nervous System”, London: J. & A.
Churchill.
Bibliography
114
Giovannoni, G.; Schalkwyk, J.; Fritz, V.U. & Lees, A.J. (1999): “Bradykinesia akinesia inco-
ordination test (BRAIN TEST): an objective computerized assessment of upper limb motor
function”, Journal of Neurol Neurosurg Psychiatry, vol. 67, pp. 624–629.
Giuffrida, J.P.; Riley, D.E.; Maddux, B.N. & Heldman, D.A. (2009): “Clinically deployable
Kinesia technology for automated tremor assessment”, Movement Disorders, vol. 24, pp.
723–730.
Hauser, R.A.; Friedlander, J.; Zesiewicz, T.A.; Adler, C.H.;Seeberger, C.H.; O’Brien C.F.; et
al. (2000): “A home diary to assess functional status in patients with Parkinson’s disease with
motor fluctuations and dyskinesia”, Clinical Neuropharmacology, vol. 23, no. 2, pp. 75–81.
Heldman, D.A.; Giuffrida, J.P.; Chen, R.; Payne, M.; Mazzella, F.; Dunn, E. et al. (2011a):
“The modified bradykinesia rating scale for Parkinson’s disease: reliability and comparison
with kinematic measures”, Movement Disorders, vol. 26, pp. 1859–1863.
Heldman, D.A.; Jankovic, J.; Vaillancourt, D.E.; Prodoehl, J.; R.J. Elble & Giuffrida, J.P.
(2011b): “Essential tremor quantification during activities of daily living”, Parkinsonism &
Related Disorders, vol. 17, no. 7, pp. 537–542.
Hess, C.W. & Pullman, S.L. (2012): “Tremor: clinical phenomenology and assessment tech-
niques”, Tremor and Other Hyperkinetic Movements, vol. 2, pp. 1–15.
Hurtado, J.M.; Charles, C.M.; Tamas, L.B. & Sigvardt, K.A. (1999): “Dynamics of tremor-
related oscillations in the human globus pallidus: A single case study”, Proc. Natl. Acad. Sci.
U.S.A, vol. 96, no. 4, pp. 1674–1679.
Iacono, R.P.; Lonser, R.R.; Maeda, G.; Kuniyoshi, S.; Warner, D.; Mandybur, G, & Yamada
S. (1995): “Chronic anterior pallidal stimulation for Parkinson’s disease”, Acta Neurochir.,
vol. 137, pp. 106–112.
Iba´nez, J.; Serrano, J.I.; del Castillo, M.D. & Barrios, L.J. (2010): “An asynchronous BMI
system for online single-Trial detection of movement intention”, IEEE EMBS Conf., pp.
4562–4565.
Jankovic, J. (2008): “Parkinson’s disease: clinical features and diagnosis”, J. Neurol.
Neurosurg Psychiatry, vol. 79, no. 79, pp. 368–376.
Kazi, S.; Azizan, A.; Zain, Z.M.D. & Mailah, M. (2010): “Performance evaluation of smart
glove applied to experimental rig to control human hand tremor for Parkinson disease”, ACE
Proceedings of the 9th WSEAS international conference on Applications of computer engi-
neering, pp. 306–315.
Keijsers, N.L.; Horstink, M.W.; van Hilten, J.J.; Hoff, J.I. & Gielen, C.C. (2000): “Detection
and assessment of the severity of Levodopa-induced dyskinesia in patients with Parkinson’s
disease by neural networks”, Movement Disorders, vol. 15, no. 6, pp. 1104–1111.
Kim, J.W.; Lee, J.H.; Kwon, Y.; Kim, C.S.; Eom, G.M.; Koh, S.B.; Kwon, D.Y. & Park,
K.W. (2011a): “Quantification of bradykinesia during clinical finger taps using a gyroscope
in patients with Parkinson’s disease”, Medical Biology Engineering Computation, vol. 49, pp.
365–371.
Bibliography
115
Kim, J.W.; Kwon, J.; Kim, Y.M.; Chung, H.Y.; Eom, G.M.; Jun, J.H.; et al. (2011b): “Analy-
sis of lower limb bradykinesia in Parkinson's disease patients”, Geriatrics & Gerontology
International, vol. 12, no. 2, pp. 257–264.
Kringelbach, M.L.; Jenkinson, N.; Owen, S.L.F. & Aziz, T.Z. (2007): “Translational principle
of deep brain stimulation”, Nature Reviews Neuroscience, vol. 8, no. 8, pp. 623–635.
Levin, J.; krafczyk, S.; Valkovic, P.; Eggert, T.; Claasssen, J. & Bötzl, K. (2009): “Objective
measurement of muscle rigidity in Parkinsonian patients treated with subthalamic stimula-
tion”, Movement Disorders, vol. 24, no. 1, pp. 57–63.
LeMoyne, R.; Mastronianni, T.; Cozza, M.; Coroian, C. & Grundfest, W. (2010): “Implemen-
tation of an iPhone for characterizing Parkinson’s disease tremor through a wireless accel-
erometer application”, IEEE EMBS Conf., pp. 4954–4958.
LeMoyne, R.; Coroian, C. & Mastronianni, T. (2009): “Quantification of Parkinson’s disease
characteristics using wireless accelerometers”, IEEE Complex Medical Engineering Conf.
(CME/ICME), pp. 1–5.
Lötters, J.C.; Schipper, J.; Veltink, P.H.; Olthuis, W. & Bergveld, P. (1998): “Procedure for
in-use calibration of triaxial accelerometers in medical applications”, Sensors and Actuators
A: Physical, vol. 68, pp. 221–228.
Luce, R.D. & Krumhansl, C. (1988): “Measurement, scaling, and psychophysics”, in Stevens'
Handbook of Experimental Psychology (Atkinson R.C., Herrnstein R.J., Lindzey G., & Luce
R.D.), New York: Wiley, pp. 1–74.
Lueth, T.C.; D’Angelo, L.T. & Czabke, A. (2010): “Chart 4: TUM–AgeTech: A New
Framework for Pervasive Medical Devices”, in Pervasive and Smart Technologies for
Healthcare: Ubiquitous Methodologies and Tools (Coronato, A. & Pietro, G.D.). New York:
IGI Global, pp. 295–321.
Luinge, H.J. & Veltink, P. H. (2005): “Measuring orientation of human body segments using
miniature gyroscopes and accelerometers”, Med. Biol. Eng. Comput., vol. 43, pp. 273–282.
Louis, E.D. & Ferreira, J.J. (2010): “How common is the most common adult movement dis-
order? Update on the worldwide prevalence of essential tremor”, Mov. Disord., vol. 25, pp.
534–541.
Machado, A.; Rezai, A.R.; Kopell, B.H.; et al. (2006): “Deep brain stimulation for Parkin-
son’s disease: surgical technique and perioperative management”, Movement Disorders, vol.
21, Suppl. 14, pp. 247–258.
Madgwick, S.O.H.; Harrison, A.J.L. & Vaidyanathan, A. (2011): “Estimation of IMU and
MARG orientation using a gradient descent algorithm”, IEEE ICORR Conf., Zurich, pp. 1–7.
Mauro, R. (1990): “Understanding L.O:V.E. (Left Out Variables Error): a method for estimat-
ing the effects of omitted variables”, Psychological Bulletin, vol. 108, no. 2, pp. 314–329.
McClelland III, S. (2011): “A cost analysis of intraoperative microelectrode recording during
subthalamic stimulation for Parkinson’s disease”, Movement Disorders, vol. 26, no. 8, pp.
1422–1427.
Bibliography
116
Mera, T.O.; Vitek, J.L.; Alberts, J.L. & Giuffrida, J.P. (2011): “Kinematic optimization of
deep brain stimulation across multiple motor symptoms in Parkinson’s disease”, J. Neurosci.
Methods, vol. 198, no. 2, pp. 280–286.
Mera, T.O.; Burack, M.A. & Giuffrida, J.P. (2012): “Quantitative assessment of Levodopa-
induced dyskinesia using automated motion sensing technology”, IEEE EMBS Conf., pp.
154–157.
Moore, G.P.; Ding, L. & Bronte-Stewart, H.M. (2000): “Concurrent Parkinson tremors”, J
Physiol., vol. 529, pp. 273–281.
Mostile, G.; Giuffrida, J.P.; Adam, O.R.; Davidson, A. & Jankovic, J. (2010): “Correlation
between Kinesia system assessments and clinical tremor scores in patients with essential
tremor movement”, Movement Disorders, vol. 25, no. 12, pp. 1938–1943.
Nakagawa, S. & Cuthill, I.C. (2010): “Effect size, confidence interval and statistical signifi-
cance: a practical guide for biologists”, Biological Reviews Cambridge Philosophical Society,
vol. 82, pp. 591–605.
Narcisa, V.; Aguilar, D.; Nquyen, D.V.; et al. (2011): “A quantitative assessment of tremor
and ataxia in female FMR1 premutation carriers using CATSYS”, Curr. Gerontol. Geriatr.
Res., vol. 2011, pp. 1–7.
National Instrument Inc. (2011): LabVIEW User Tutorials. USA.
NDI Digital Inc. (2010): Aurora V2 User Guide. Canada.
Niazmand, K.; Jehle, C.; D’Angelo, L.T. & Lueth, T.C. (2010): “A new washable low-cost
garment for everyday fall detection”, IEEE EMBS Conf., Buenos Aires, pp. 6377–6380.
Niazmand, K.; Tonn, K.; Kalaras, A.; Fietzek, U.M.; Mehrkens, J.H. & Lueth, T.C. (2011a): “
Quantitative evaluation of Parkinson's disease using sensor based smart glove”, IEEE Sympo-
sium on Computer-Based Medical Systems-CBMS, Bristol, pp. 1–8.
Niazmand, K., Kalaras, A., Dai, H., Lueth, T.C. (2011b): “Comparison of methods for tremor
frequency analysis for patients with Parkinson’s disease”, The 4th International Conference
on BioMedical Engineering and Informatics (IEEE CBEI), Shanghai, pp: 1312-1316.
Norman, K.E.; Edwards, R. & Beuter, A. (1999): “The measurement of tremor using a veloci-
ty transducer: comparison to simultaneous recordings using transducers of displacement, ac-
celeration and muscle activity”, Journal of Neuroscience Methods, vol. 92, pp. 41–54.
Okun, M.S. & Foote, K.D. (2010): “Parkinson’s disease DBS: what, when, who and why?
The time has come to tailor DBS targets”, Expert Review Neurotherapeutics, vol. 10, pp.
1847–1857.
Okuno, R.; Yokoe, M.; Akazawa, K.; Abe, K. & Sakoda, S. (2006): “Finger taps movement
acceleration measurement system for quantitative diagnosis of Parkinson’s disease”, IEEE
EMBS Conf., Suppl, pp. 6623–6626.
Olivares, A.; Olivares, G.; Gorriz, J.M. & Ramirez, J. (2009): “High-efficiency low-cost ac-
celerometer-aided gyroscope calibration”, IEEE International Conference on Test and Meas-
urement (ICTM), pp. 354–360.
Bibliography
117
Oluigbo, C.O.; Salma, A. & Rezai, A.R. (2012): “Deep brain stimulation for neurological
disorders”, IEEE Reviews in Biomedical Engineering, vol. 5, pp. 88–99.
Ostrem, J.L.; Galifianakis, N.B.; Markun, L.C.; Grace, J.K.; Martin, A.J.; Starr, P.A. & Lar-
son PS. (2013): “Clinical outcomes of PD patients having bilateral STN DBS using high-field
interventional MR-imaging for lead placement”, Clin. Neurol. Neurosurg., vol. 115, no. 6, pp.
708–712.
O'Suilleabhain, P.E. & Matsumoto, J.Y. (1998): “Time-frequency analysis of tremors”, Brain,
vol. 121, pp. 2127–2134.
Park, B.K.; Kwon, Y.; Kim, J.W.; Lee, J.H.; Eom, G.M.; Koh, S.B. et al. (2011): “Analysis of
viscoelastic properties of wrist joint for quantification of parkinsonian rigidity”, IEEE Trans-
actions on Neural Systems and Rehabilitation Engineering, vol. 19, no. 2, pp. 167–176.
Patel, S.; Hughes, R.; Huggins, N.; Standaert, D.; Growdon, J.; Dy, J. & Bonato, P. (2008):
“Using wearable sensors to predict the severity of symptoms and motor complications in late
stage Parkinson's disease”, IEEE EMBS Conf., pp. 3686–3689.
Patel, S.; Lorincz, K.; Hughes, R.; Huggins, N.; Growdon, J.; Standaert, D. et al. (2009):
“Monitoring motor fluctuations in patients with Parkinson’s disease using wearable sensors”,
IEEE Trans. Inf. Technol. in Biomed., vol. 13, no. 6, pp. 864–873.
Patrick, S.K.; Denington, A.A.; Gauthier, M.J.; Gillard, D.M. & Prochazka, A. (2001):
“Quantification of the UPDRS rigidity scale”, IEEE Trans. Neural Syst. Rehabil. Eng., vol. 9,
pp. 31–41.
Pahwa, R. & Lyons, K.E. (2007): Handbook of Parkinson’s Disease (4th ed.), New York:
Informa Healthcare (USA) Inc.
Pavel, P. & Popelka, J. (2012): “IMU aiding using two AHRS units”, IEEE/AIAA Digital Avi-
onics System Conference (DASC), pp. 5B1-1–5B1-13.
Post, B.; Maruschka, M.P.; de Bie, R.M.; de Haan, R.J. & Speelman, J.D. (2005): “Unified
Parkinson's disease rating scale motor examination: are ratings of nurses, residents in neurol-
ogy, and movement disorders specialists interchangeable?”, European Journal of Nuclear
Medicine and Molecular Imaging, vol. 20, no. 12, pp. 1577–1584.
Powell, Jr.H.C.; Hanson, M.A. & Lach, J. (2007): “A wearable inertial sensing technology for
clinical assessment of tremor”, IEEE Biomedical Circuits and Systems Conference, BIOCAS
2007, pp. 9–12.
Popovic, L.Z., Sekara, T.B. & Popovic, M.B. (2008): “Adaptive band-pass filter (ABPF) for
tremor extraction from inertial sensor data”, Computer Methods and Programs in Biomedi-
cine, vol. 99, pp. 298–305.
Prochazka, A.; Bennett, D.J.; Stephens, M.J.; Patrick, S.K.; Sears-Duru, R.; Roberts, T. &
Jhamandas J.H. (1997): “Measurement of rigidity in Parkinson’s disease”, Movement Disor-
ders, vol. 12, pp. 24–32.
Bibliography
118
Richard, G.B.; Sasha, C.B.; Peter, G.B.; Sarah, L.O.; Carol, J.; David, S. & Tipu, Z.A. (2005):
“Deep brain stimulation for movement disorders and pain”, Journal of Clinical Neuroscience,
vol. 12, no. 4, pp. 457–463.
Rigas, G.; Tzallas, A.T.; Tsalikakis, D.G.; Konitsiotis, S. & Fotiadis, D.I. (2009) “Real-time
quantification of rest tremor in the Parkinson’s disease”, IEEE EMBS Conf., Minneapolis, pp.
1306-1309.
Rissanen, S.M.; Kankannpää, M.; Tarvainen, M.P.; Nuutinen, J.; Tarkka, I.M.; Airaksinen, O.
& Karialainen, P.A. (2007): “Analysis of surface EMG signal morphology in Parkinson’s
disease”, Physiological Measurement, vol. 28, pp. 1507–1521.
Rissanen, S.M.; Kankaanpaa, M.; Tarvainen, M. P.; Novak, V.; Hu, K.; Manor, B. et al.
(2011): “Analysis of EMG and acceleration signals for quantifying the effects of Deep Brain
Stimulation in Parkinson’s disease”, IEEE Trans. Biomed. Eng., vol. 99, pp. 1–9.
Riviere, C.N.; Reich, S.G. & Thankor, N.V. (1997): “Adaptive Fourier modelling for quanti-
fication of tremor”, Journal of Neuroscience Methods, vol. 74, pp: 77–87.
Sakoda, S.; Akazawa, K.; Okuno, R.; Yokoe, M. & Endo, T. (2011): “Muscle tonus measur-
ing apparatus”, United States Patent Application Publication, pub. no.: US2011/0087128 A1.
Salarian, A.; Russmann, H.; Wider, C.; Burkhard, P.R.; Vingerhoets, F.J. & Aminian, K.
(2007): “Quantification of tremor and bradykinesia in Parkonson’s disease using a novel am-
bulatory monitoring system”, IEEE Trans. Biomed Eng., vol. 54, pp. 313–322.
Sepehri, B.; Esteki, A.; Ebrahimi-Takamjani, E. et al. (2007): “Quantification of rigidity in
Parkinson’s disease”, Ann. Biomed. Eng., vol. 35, pp. 2196–2203.
Singh, A.; Kammermeier, S.; Mehrkens, J.H. & Bötzel, K. (2012): “Movement kinematic
after deep brain stimulation associated microlesions”, J. Neurol. Neurosurg. Psychiatry, vol.
82, pp. 1022–1026.
Shaeffer, D.K. (2013): “MEMS inertial sensors: A tutorial overview”, IEEE Communications
Magazine, vol. 51, no.4, pp. 100–109.
Shany, T.; Redmond, S.J.; Narayanan, M.R. & Lovell, N.H. (2012): “Sensors-based wearable
systems for monitoring of human movement and falls”, IEEE Sensors Journal, vol. 12, no. 3,
pp. 658–670.
Shapiro, M.B.; Vaillancourt, D.E.; Sturman, M.M.; et al. (2007): “Effects of STN DBS on
rigidity in Parkinson’s disease”, IEEE Trans. Neural. Syst. Rehabil. Eng., vol. 15, pp. 173–
181.
Shima, K.; Tsuji, T.; Kan, E.; Kandori, A., Yokoe, M. & Sakoda, S. (2008): “Measurement
and evaluation of finger tapping movements using magnetic sensors”, IEEE EMBS Conf., pp.
5628–5631.
Smeja, M.; Foerster, F.; Fuchs, G.; Emman, D.; Hornig, A. & Fahrenberg, J. (1999): “24-h
assessment of tremor activity and posture in Parkinson’s disease by multi-channel
acceelerometry”, Journal of Psychosiology, vol. 13, pp. 245–256.
Bibliography
119
Spieker, S.; Jentgens, C.; Boose, A. & Dichgans, J. (1995): “Reliability, specificity and sensi-
tivity of long-term tremor recordings”, Electroencephalography and Clinical Neurophysiolo-
gy, vol. 97, pp. 326–331.
Spyers-Ashby, J.M. & Stokes, M.J. (2000): “Reliability of tremor measurements using a mul-
tidimensional electromagnetic sensor system”, Clin. Rehabil., vol. 14, no. 4, pp. 425–432.
Sturman, M.M.; Vaillancourt, D.E.; Metman, L.V.; Bakay, R.A.E & Corcos, D.M. (2004):
“Effects of subthalamic nucleus stimulation and medication on rest and postural tremor in
Parkinson’s disease”, Brain, vol. 127, pp. 2131–2143.
Su, Y.; Allen, C.R.; Geng, D.; Burn, D.; Brechany, U.; Bel, G.D. & Rowland, R.(2003): “3-D
motion system ("data-gloves"): application for Parkinson's”, IEEE Trans. Instr. Measur., vol.
52, no. 3, pp. 662–674.
Synnott, J.; Chen, L.; Nugent, C.D. & Moore, G. (2012): “WiiPD-objective home assessment
of Parkinson’s disease using the Nintendo Wii remote”, IEEE Trans. Inf. Technol. in Biomed.,
vol. 16, no. 6, pp. 1304–1312.
Tagliati, M.; Guten, G.N. & Horne, J. (2007): “Parkinson’s Disease for Dummies”, Indianap-
olis: Wiley Publishing Inc.
Takuyuki, E.; Masaru, Y.; Harutoshi, F. & Saburo, S. (2011): “Chapter 9: Novel Methods to
Evaluate Symptoms in Parkinson’s Disease–Rigidity and Finger Tapping”, in Diagnostics and
Rehabilitation of Parkinson’s Disease (Dushanova, J.), Croatia: InTech, pp. 191–206.
Teräväinen, H. & Calne, D.B. (1980): “Action tremor in Parkinson's disease”, J. Neurol.
Neurosurg. Psychiatry, vol. 43, no. 3, pp. 257–267.
Timmer, J.; Gantert, C.; Deuschl G. & Honerkamp, J. (1993): “Characteristics of hand tremor
time series”, Biol. Cybern., vol. 70, pp. 75–80.
Timmer, J.; Lauk, M. & Lücking, C.H. (1997): “Confidence regions for spectral peak fre-
quencies”, Biometrical Journal, vol. 39, no. 7, pp. 849–861.
Timmer, J.; Lauk, M.; Vach, W. & Lücking, C.H. (1999): “A test for a difference between
spectral peak frequencies”, Journal Computational Statistics & Data Analysis, vol. 30, no. 1,
pp. 45–55.
Timmer, J.; Häussler, S.; Lauk, M. & Lücking, C.H. (2000a): “Pathological tremors: deter-
ministic chaos or nonlinear stochastic oscillators?”, Chaos, vol. 10, no. 1, pp. 278–288.
Timmer, J.; Lauk, M.; Häußler, S. & Radt, V. (2000b): “Cross-spectral analysis of tremor
time series”, International Journal of Bifurcation and Chaos, vol. 10, no. 11, pp. 2595–2610.
Thümler, R. (2002): “Morbus Parkinson: Ein Leitfaden für Klinik und Praxis”, Berlin:
Springer.
Van Dillen, L.R.; Roach, K.E. (1988): “Interrater Reliability of a Clinical Scale of Rigidity”,
Physical Therapy, vol. 68, pp. 1679–1681.
Bibliography
120
Van Someren, E.J.; Vonk, B.E.; Thijssen, W.A.; Speelman, J.D.; Schuurman, P.R.; Mirmiran,
M. & Swaab, D.F. (1998): “A new actigraph for long-term registration of the duration and
intensity of tremor and movement”, IEEE Trans. Biomed. Eng., vol. 45, no. 3, pp. 386–395.
Veluvolu K.C. & Ang, W.T. (2011): “Estimation of physiological tremor from accelerometers
for real-time applications”, Sensors, vol. 11, no. 3, pp. 3020–3036.
Winestone, J.S.; Zaidel, A.; Bergman, H. & Israel, Z. (2006): “The use of macroelectordes in
recording cellular spiking activity”, Clinical Neuroscience, vol. 206, pp. 34–39.
Winqeier, B.; Tcheng, T.; Koop, M.M.; Hill, B.C.; Heit, G. & Bronte-Stewart, H.M. (2006):
“Intra-operative STN DBS attenuates the prominent beta rhythm in the STN in Parkinson’s
disease”, Exp. Neurol., vol. 197, no. 1, pp. 244–251.
Yuting, Z.; Markovic, S; Wagenaar, R.C. & Little, T.D.C. (2011): “Continuous functional
activity monitoring based on wearable tri-axial accelerometer and gyroscope”, Pervasive
Computing Technologies for Healthcare (PervasiveHealth), Dublin, vol. 99, pp. 1–9.